# forck.live — 11–17 May 2026

> AI lab announcements, news reports and podcasts.

forck.live follows what companies announce about their own products. Press may
corroborate an official story; it never leads and never appears as an item of its
own. A missing date stays missing. A title is the publisher's words — where a
company does not publish in English, an English rendering is listed separately
as `Title in English` and named as this site's translation, never as theirs.

This file is one issue of the weekly archive: every confirmed
first-party announcement of ISO week 2026-W20, 11–17 May 2026, newest
first. The window never moves, so this URL always names the same seven days —
but a post published inside them and detected later is added when it arrives,
so a closed week can still gain entries.

Every entry is one first-party announcement, and carries the same eight fields
in this order before any others:

    Company · Announced · Category · Coverage · Announcement · Group · Source · Record

`Source` is the company's own page the announcement was traced to and `Record`
is this site's page for it. Both are safe to cite.

`Coverage` is a number of outlets, or `not counted` — which means this record
reached no count of its own, never that nobody wrote about it.

`Category` is this tracker's own analysis, and is authoritative wherever it is
stated. `not stated` means either that nothing has analysed the post yet or
that the analysis placed it outside this taxonomy; neither is a verdict about
the announcement.

`Announcement` is whether the post is news of something the company did, rather
than a how-to, a customer story, a changelog listing or a staffing note.

`Group` is which of this record's groups the entry falls in, carried on the
entry itself so it holds where the groups are headings and where they are not.
One of: `models`, `covered`, `announcements`, `routine`.

Anything after `Record` is held detail that only some announcements carry.

- This week's page: https://forck.live/week/2026-W20
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Google DeepMind — Simulate real-world places with Project Genie and Street View

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-17T19:53:18+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/simulate-real-world-places-with-project-genie-and-street-view/
- Record: https://forck.live/items/1221-simulate-real-world-places-with-project-genie-and-street-view
- Subject: Gemini

Google DeepMind is expanding access to Google AI Ultra subscribers globally and introducing a new capability powered by Street View.

### Google DeepMind — Gemini for Science: AI experiments and tools for a new era of discovery

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-17T13:50:34+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/gemini-for-science-ai-experiments-and-tools-for-a-new-era-of-discovery/
- Record: https://forck.live/items/1224-gemini-for-science-ai-experiments-and-tools-for-a-new-era-of-discovery
- Subject: Gemini

A collection of science tools and experiments aimed at expanding the scale and precision of scientific exploration.

### Google DeepMind — Making it easier to understand how content was created and edited

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-17T13:43:50+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/making-it-easier-to-understand-how-content-was-created-and-edited/
- Record: https://forck.live/items/1225-making-it-easier-to-understand-how-content-was-created-and-edited
- Subject: Gemini

Google DeepMind is expanding its tools to help users understand how content was created and edited across the web.

### Genspark — The Last Three Things AI Can't Copy

- Company: Genspark (genspark.ai)
- Announced: 2026-05-17
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/seeing-agi-impact9
- Record: https://forck.live/items/18254-the-last-three-things-ai-can-t-copy
- Subject: Genspark

### Google DeepMind — Strengthening Singapore’s AI Future: A New National Partnership

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-16T09:13:34+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/strengthening-singapores-ai-future-a-new-national-partnership/
- Record: https://forck.live/items/1226-strengthening-singapore-s-ai-future-a-new-national-partnership
- Subject: Gemini

Google DeepMind and Singapore have formed a partnership to apply frontier AI to complex challenges in health, education, and sustainability.

### Google DeepMind — Finding the molecular switches behind new infectious diseases

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-16T08:16:06+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/finding-the-molecular-switches-behind-new-infectious-diseases/
- Record: https://forck.live/items/1227-finding-the-molecular-switches-behind-new-infectious-diseases
- Subject: Gemini

Clare Bryant uses Co-Scientist to identify genetic triggers in emerging infectious diseases, as described in a Google DeepMind blog post. The post does not announce a new model or product launch; it is a research application story.

### Google DeepMind — Opening new paths in aging research

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-16T08:08:44+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/opening-new-paths-in-aging-research/
- Record: https://forck.live/items/1228-opening-new-paths-in-aging-research
- Subject: Gemini

Calico Life Sciences uses the Co-Scientist tool to connect scattered findings and generate new leads in aging research.

### Google DeepMind — Accelerating discovery of liver disease mechanisms

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-16T08:00:15+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/accelerating-discovery-of-liver-disease-mechanisms/
- Record: https://forck.live/items/1229-accelerating-discovery-of-liver-disease-mechanisms
- Subject: Gemini

Filippo Menolascina uses Co-Scientist to identify new liver disease treatments and explain why existing drugs only help certain patients.

### Google DeepMind — Uniting biological toolkits for a new approach to ALS

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-16T07:53:11+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/uniting-biological-toolkits-for-a-new-approach-to-als/
- Record: https://forck.live/items/1230-uniting-biological-toolkits-for-a-new-approach-to-als
- Subject: Gemini

Google DeepMind's Co-Scientist unites Boston Children's Hospital and MIT's labs to explore new RNA-based treatments for ALS.

### Google DeepMind — Uncovering repurposed medicines to fight liver fibrosis

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-16T07:40:27+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/uncovering-repurposed-medicines-to-fight-liver-fibrosis/
- Record: https://forck.live/items/1231-uncovering-repurposed-medicines-to-fight-liver-fibrosis
- Subject: Gemini

A Stanford geneticist uses the Co-Scientist system to help find new treatments for chronic liver disease and liver fibrosis.

### Google DeepMind — How WeatherNext helped the National Hurricane Center better predict Hurricane Melissa’s historic landfall in Jamaica

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-16T03:14:17+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/how-weathernext-helped-the-national-hurricane-center-better-predict-hurricane-melissas-historic-landfall-in-jamaica/
- Record: https://forck.live/items/1232-how-weathernext-helped-the-national-hurricane-center-better-predict-hurricane
- Subject: Gemini

The WeatherNext AI model helped the National Hurricane Center predict Hurricane Melissa's landfall, giving communities more preparation time.

### Krea — Style references in Krea 2

- Company: Krea (krea.ai)
- Announced: 2026-05-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/style-references-krea-2
- Record: https://forck.live/items/15107-style-references-in-krea-2
- Subject: Krea 2 / AI creative platform

Krea's guide explains how style references (S-Refs) work in Krea 2, including the strength slider that controls how much influence a reference image has on the output, and when to use S-Refs versus mood boards.

### OpenAI — OpenAI and Malta partner to bring ChatGPT Plus to all citizens

- Company: OpenAI (openai.com)
- Announced: 2026-05-16
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/malta-chatgpt-plus-partnership
- Record: https://forck.live/items/121-openai-and-malta-partner-to-bring-chatgpt-plus-to-all-citizens
- Subject: GPT / ChatGPT / API

OpenAI and Malta partner to offer ChatGPT Plus and training to all citizens, expanding AI access.

### Google DeepMind — Gemini 3.5: frontier intelligence with action

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-15T22:50:12+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://deepmind.google/blog/gemini-3-5-frontier-intelligence-with-action/
- Record: https://forck.live/items/1233-gemini-3-5-frontier-intelligence-with-action
- Subject: Gemini

Google DeepMind announces Gemini 3.5, a model designed to support complex and agentic workflows.

### fal — Building long-term trust in a world where creation moves at the speed of thought

- Company: fal (fal.ai)
- Announced: 2026-05-15T18:58:48+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.fal.ai/building-long-term-trust-in-a-world-where-creation-moves-at-the-speed-of-thought/
- Record: https://forck.live/items/7998-building-long-term-trust-in-a-world-where-creation-moves-at-the-speed-of-thought
- Subject: fal platform

fal, an AI infrastructure company, detailed its trust and safety posture, including proactive integration of Thorn for CSAM detection and a partnership with StopNCII.org for non-consensual intimate imagery hash-matching. The company stated it operates on a principle of acting immediately upon knowledge of a violation and frames its responsibility as closer to a cloud provider than a content platform.

### Factory AI — Factory acquires Lumetric

- Company: Factory AI (factory.com)
- Announced: 2026-05-15
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/lumetric
- Record: https://forck.live/items/17836-factory-acquires-lumetric
- Subject: Droid

Factory has acquired Lumetric, a YC W24 team that builds model-agnostic AI systems for high-context workflows. The acquisition aims to deepen Factory's investment in autonomous systems for software engineering, integrating Lumetric's experience in applied AI products into Factory's platform.

### OpenClaw — Where OpenClaw Security Is Heading

- Company: OpenClaw (openclaw.ai)
- Announced: 2026-05-15
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openclaw.ai/blog/where-openclaw-security-is-heading/
- Record: https://forck.live/items/17766-where-openclaw-security-is-heading
- Subject: OpenClaw

Our goal is for OpenClaw to become a trusted way to run a powerful AI personal assistant. OpenClaw can read files, run commands, install plugins, talk to the network, and act on a real machine for a real user.

### Krea — Krea at Cannes Film Festival

- Company: Krea (krea.ai)
- Announced: 2026-05-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.krea.ai/blog/krea-at-cannes-film-festival
- Record: https://forck.live/items/15109-krea-at-cannes-film-festival
- Subject: Krea 2 / AI creative platform

Krea is attending the Cannes Film Festival, where its co-founder and CTO will demonstrate real-time visual generation on stage at the American Pavilion on May 17.

### Krea — Krea 2 vs Midjourney: which AI image generator is right for you?

- Company: Krea (krea.ai)
- Announced: 2026-05-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.krea.ai/blog/krea-2-vs-midjourney
- Record: https://forck.live/items/15108-krea-2-vs-midjourney-which-ai-image-generator-is-right-for-you
- Subject: Krea 2 / AI creative platform

Krea compares its new Krea 2 foundation image model to Midjourney, highlighting Krea 2's style control tools (style references with strength sliders and mood boards) and real-time canvas workflow versus Midjourney's queue-based generation. Krea 2 is part of an integrated suite including video, editing, and LoRA training, with a free tier and paid plans starting at $9/month.

### Krea — Mood boards in Krea 2

- Company: Krea (krea.ai)
- Announced: 2026-05-15
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/moodboards-krea-2
- Record: https://forck.live/items/15080-mood-boards-in-krea-2
- Subject: Krea 2 / AI creative platform

Krea introduced mood boards in Krea 2, a tool that accepts more than four reference images and uses custom LLMs and clustering to capture concepts, expressions, characters, and overall vibe—not just style. Users analyze a board to generate a taste profile, keywords, and avoids, then prompt the image tool to produce outputs that reflect the full aesthetic of the reference set. The feature also ships with curated presets such as Retro web, Expressive marker, and Film noir.

### OpenAI — Operations workflows with ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-05-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/chatgpt-work/how-business-operations-teams-use-codex
- Record: https://forck.live/items/4146-operations-workflows-with-chatgpt-work
- Subject: GPT / ChatGPT / API

OpenAI Academy published a guide on how business operations teams can use ChatGPT Work to consolidate scattered project context, metrics, and stakeholder input into decision-ready briefs, updates, packets, and tradeoff models.

### OpenAI — How business operations teams use ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-05-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/chatgpt-work/how-business-operations-teams-use-codex
- Record: https://forck.live/items/3714-how-business-operations-teams-use-chatgpt-work
- Subject: GPT / ChatGPT / API

OpenAI Academy published a guide on how business operations teams can use ChatGPT Work to consolidate scattered project information and produce decision-ready documents such as briefs, updates, and tradeoff models, with a note that the webinar was recorded using the former Codex app but can now be followed in ChatGPT Work at chatgpt.com or the desktop app.

### xAI — Connect Grok to Hermes Agent

- Company: xAI (x.ai)
- Announced: 2026-05-15
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://x.ai/news/grok-hermes
- Record: https://forck.live/items/3467-connect-grok-to-hermes-agent
- Subject: Grok

Grok can now be used inside Nous Research's open-source, self-improving Hermes agent with a Grok account and subscription.

### Together AI — Together AI and Pearl Research Labs Team Up to Reduce the Cost of AI Inference

- Company: Together AI (together.ai)
- Announced: 2026-05-15
- Category: api-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/together-ai-partners-with-pearl-research-labs
- Record: https://forck.live/items/2346-together-ai-and-pearl-research-labs-team-up-to-reduce-the-cost-of-ai-inference
- Subject: Inference platform

Together AI partners with Pearl Research Labs to offer a discounted inference endpoint for the Gemma-4-31B-it-pearl model, using Pearl's blockchain protocol to offset costs via crypto emissions. The endpoint is discounted by over 25%, offset by the future value of crypto emissions, with the expectation that further price reductions may follow as the coin value increases. This is the first product integrating the Pearl Network, which uses Proof of Useful Work to simultaneously generate cryptocurrency while performing AI inference. The partnership aims to reduce inference costs by leveraging the economic throughput of GPUs to produce a digital asset at no additional cost. Together AI plans to expand its Pearl-powered product portfolio and offer customers a direct portion of the currency emissions in the future.

### OpenAI — A new personal finance experience in ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2026-05-15
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/personal-finance-chatgpt
- Record: https://forck.live/items/124-a-new-personal-finance-experience-in-chatgpt
- Subject: GPT / ChatGPT / API

OpenAI is previewing a new personal finance experience within ChatGPT, available for Pro users in the U.S. It allows users to securely connect financial accounts and receive AI-powered insights and guidance.

### OpenAI — Databricks brings GPT-5.5 to enterprise agent workflows

- Company: OpenAI (openai.com)
- Announced: 2026-05-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/databricks
- Record: https://forck.live/items/123-databricks-brings-gpt-5-5-to-enterprise-agent-workflows
- Subject: GPT / ChatGPT / API

Databricks uses GPT-5.5 for enterprise agent workflows after the model set a new state of the art on the OfficeQA Pro benchmark.

### OpenAI — How business operations teams use ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-05-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/codex-for-work/how-business-operations-teams-use-codex
- Record: https://forck.live/items/122-how-business-operations-teams-use-chatgpt-work
- Subject: GPT / ChatGPT / API

Business operations teams can use ChatGPT Work to produce initiative briefs, strategy updates, leadership decision packets, and progress updates from real work inputs.

### OpenAI — Sea's View on the Future of Agentic Software Development with Codex

- Company: OpenAI (openai.com)
- Announced: 2026-05-14T20:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/sea-david-chen
- Record: https://forck.live/items/125-sea-s-view-on-the-future-of-agentic-software-development-with-codex
- Subject: GPT / ChatGPT / API

Sea Limited's CPO explains why the company is deploying Codex across engineering teams to accelerate AI-native software development in Asia.

### Hugging Face — Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — Best Sub-100M Retrieval Quality

- Company: Hugging Face (huggingface.co)
- Announced: 2026-05-14T18:55:01+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/ibm-granite/granite-embedding-multilingual-r2
- Record: https://forck.live/items/1503-granite-embedding-multilingual-r2-open-apache-2-0-multilingual-embeddings-with
- Subject: Platform

Hugging Face announces the release of two new multilingual embedding models from IBM: a 311M-parameter full-size model and a 97M-parameter compact model, both under Apache 2.0 license, supporting 200+ languages with 32K token context and improved retrieval quality.

### Harvey — Building an Agentic Security Operations Center

- Company: Harvey (harvey.ai)
- Announced: 2026-05-14T17:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/building-an-agentic-security-operations-center
- Record: https://forck.live/items/17565-building-an-agentic-security-operations-center
- Subject: Harvey

Harvey describes its agentic security operations center (SOC), a system of always-on AI agents that hunt, triage, investigate, and author detections, built on a machine-readable world model of the company's threat surface and gated by human review on every production change.

### Harvey — The Brief: May 2026

- Company: Harvey (harvey.ai)
- Announced: 2026-05-14T15:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/the-brief-may-2026
- Record: https://forck.live/items/17566-the-brief-may-2026
- Subject: Harvey

Harvey expanded its platform with new document editing capabilities (PowerPoint, Excel, PDF), added Opus 4.7 and GPT-5.5 models, and introduced collaboration features including Guest Accounts and Space Analytics.

### Harvey — Preparing Law Students for AI-Powered Legal Practice

- Company: Harvey (harvey.ai)
- Announced: 2026-05-14T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harvey-academy-law-schools-course
- Record: https://forck.live/items/17567-preparing-law-students-for-ai-powered-legal-practice
- Subject: Harvey

Harvey launched a free course, Harvey Law Schools: Preparing for Legal Work, for law students in its Law School Program to build practical AI fluency covering use cases like contract drafting and M&A due diligence, with a certificate of completion. The course is available to students and faculty at more than 40 partner schools worldwide.

### OpenAI — Work with Codex from anywhere

- Company: OpenAI (openai.com)
- Announced: 2026-05-14T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/work-with-codex-from-anywhere
- Record: https://forck.live/items/126-work-with-codex-from-anywhere
- Subject: GPT / ChatGPT / API

The announcement enables users to use Codex via the ChatGPT mobile app, allowing real-time monitoring, steering, and approval of coding tasks across devices and remote environments.

### Black Forest Labs — FLUX Outpainting: Extend any image, in any direction

- Company: Black Forest Labs (bfl.ai)
- Announced: 2026-05-14T12:30:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://bfl.ai/blog/outpainting-extend-any-image-in-any-direction
- Record: https://forck.live/items/3392-flux-outpainting-extend-any-image-in-any-direction
- Subject: FLUX

Black Forest Labs announces FLUX Outpainting, a feature that allows extending any image in any direction.

### Runway — Runway is Coming to Japan

- Company: Runway (runway.com)
- Announced: 2026-05-14T10:10:10.110000+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/runway-is-coming-to-japan
- Record: https://forck.live/items/7881-runway-is-coming-to-japan
- Subject: Gen / Aleph

Runway announced its expansion into Japan, opening a Tokyo office as its Japan headquarters, committing an initial $40 million, and hiring a Head of Japan. Japan is Runway's third-largest market and fastest-growing self-serve market in Asia, with 300% growth in enterprise customers over the past 12 months.

### OpenClaw — OpenAI Models in OpenClaw, Done Right

- Company: OpenClaw (openclaw.ai)
- Announced: 2026-05-14
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openclaw.ai/blog/openai-models-in-openclaw-done-right/
- Record: https://forck.live/items/17767-openai-models-in-openclaw-done-right
- Subject: OpenClaw

OpenClaw now routes OpenAI model agent turns through the native Codex app-server runtime by default. Codex handles the model loop, tool continuation, and dynamic tool search, while OpenClaw manages channels, persona, memory, and other product features. This reduces translation overhead and improves tool handling for OpenAI models.

### Krea — On being the underdog

- Company: Krea (krea.ai)
- Announced: 2026-05-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.krea.ai/blog/on-being-the-underdog
- Record: https://forck.live/items/15110-on-being-the-underdog
- Subject: Krea 2 / AI creative platform

Krea, a 37-person company with $83M raised, released its first in-house model, Krea 2, which is designed to produce varied, artistic images rather than safe, photorealistic outputs. The model was trained on a hand-built dataset and allows users to drop reference images, adjust style influence, and build mood boards. Krea plans to train two more models on a $20M GPU cluster, with K2 described as the conservative model in the roadmap.

### Amp — npm Package Changes

- Company: Amp (ampcode.com)
- Announced: 2026-05-14
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/npm-package-changes
- Record: https://forck.live/items/7748-npm-package-changes
- Subject: Amp

Amp is shipping its CLI as a single-file executable compiled by Bun instead of as a JavaScript source package, and has renamed its npm packages: the CLI is now @ampcode/cli and the TypeScript SDK is now @ampcode/sdk, with the old names aliased until June 15, 2026.

### LG AI Research — Toward a reliable AI forecasting system: Long-term lithium price forecasting

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-05-14
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=647
- Record: https://forck.live/items/4750-toward-a-reliable-ai-forecasting-system-long-term-lithium-price-forecasting
- Subject: EXAONE

LG AI Research and LG Energy Solution jointly developed a long-term lithium price forecasting system that focuses on explainability and trustworthiness for investment decisions, addressing the limitations of traditional forecasts.

### Together AI — Violin: An open-source video translation skill that breaks language barriers

- Company: Together AI (together.ai)
- Announced: 2026-05-14
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/violin-open-source-translation-skill
- Record: https://forck.live/items/2347-violin-an-open-source-video-translation-skill-that-breaks-language-barriers
- Subject: Inference platform

Together AI announces Violin, a fully open-source video translation tool that uses ASR, LLM translation, and TTS to translate video content, with additional features like a video-content-aware chat assistant and natural language voice picker. The tool is available as a web app, CLI, and agent skills, and the codebase is released under the MIT license.

### Hugging Face — Unlocking asynchronicity in continuous batching

- Company: Hugging Face (huggingface.co)
- Announced: 2026-05-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/continuous_async
- Record: https://forck.live/items/1504-unlocking-asynchronicity-in-continuous-batching
- Subject: Platform

Asynchronous batching for continuous batching in LLM inference, which separates CPU and GPU workloads to improve GPU utilization, and has been implemented in the transformers library.

### OpenAI — Helping ChatGPT better recognize context in sensitive conversations

- Company: OpenAI (openai.com)
- Announced: 2026-05-14
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/chatgpt-recognize-context-in-sensitive-conversations
- Record: https://forck.live/items/127-helping-chatgpt-better-recognize-context-in-sensitive-conversations
- Subject: GPT / ChatGPT / API

OpenAI announces safety updates for ChatGPT that improve context awareness in sensitive conversations, enabling better detection of risk over time and safer responses.

### Lovable — Building is just the beginning: Introducing Discoverability

- Company: Lovable (lovable.dev)
- Announced: 2026-05-13T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/building-is-just-the-beginning-introducing-discoverability
- Record: https://forck.live/items/12083-building-is-just-the-beginning-introducing-discoverability
- Subject: Lovable / AI app builder

Lovable introduced discoverability capabilities within its product, making all new and existing apps readable to search engines and LLM crawlers via pre-rendered HTML. The platform also integrated Semrush for search performance insights and added an SEO dashboard for automated reviews and fixes.

### OpenAI — Building a safe, effective sandbox to enable Codex on Windows

- Company: OpenAI (openai.com)
- Announced: 2026-05-13T11:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/building-codex-windows-sandbox
- Record: https://forck.live/items/128-building-a-safe-effective-sandbox-to-enable-codex-on-windows
- Subject: GPT / ChatGPT / API

OpenAI built a sandbox to enable Codex on Windows with controlled file access and network limits.

### Runway — Introducing Runway Agent

- Company: Runway (runway.com)
- Announced: 2026-05-13T10:00:00+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/introducing-runway-agent
- Record: https://forck.live/items/7882-introducing-runway-agent
- Subject: Gen / Aleph

Runway introduced Runway Agent, an agentic creative partner that generates finished videos from a single conversation. It is designed for brand teams, marketers, creative agencies, and filmmakers, and is available now.

### Lovable — Setting the standard for agentic development

- Company: Lovable (lovable.dev)
- Announced: 2026-05-13T08:44:33+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/setting-the-standard-for-agentic-development
- Record: https://forck.live/items/12084-setting-the-standard-for-agentic-development
- Subject: Lovable / AI app builder

Lovable introduces AIUC-1, a standard for AI agent security, safety, and reliability, extended as the first certification for AI coding agents. Lovable plans to pursue AIUC-1 certification with a third-party audit scheduled for summer 2026.

### Gradium — Gradium #1 on Coval TTS Benchmarks

- Company: Gradium (gradium.ai)
- Announced: 2026-05-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://gradium.ai/blog/coval-tts-benchmarks-may-2026
- Record: https://forck.live/items/18319-gradium-1-on-coval-tts-benchmarks
- Subject: Gradium TTS / Phonon

Gradium ranks first on Coval's TTS benchmarks as of May 13, 2026, achieving the best P50 TTFA, latency range, and Word Error Rate among tested providers including ElevenLabs, Cartesia, Rime, and OpenAI. Coval is an independent evaluation platform that applies standardized, open-source methodology across providers. Gradium attributes its performance to its audio language model architecture based on Delayed Streams Modeling.

### Cognition — Devin now supports Android emulators

- Company: Cognition (cognition.com)
- Announced: 2026-05-13
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/android-emulator
- Record: https://forck.live/items/7661-devin-now-supports-android-emulators
- Subject: Devin / SWE

Devin can now spin up an Android Virtual Device (AVD), enabling autonomous development for Android applications including building, running, testing, and verifying changes in the local Android environment.

### Cursor — Development environments for cloud agents

- Company: Cursor (cursor.com)
- Announced: 2026-05-13
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/05-13-26
- Record: https://forck.live/items/2609-development-environments-for-cloud-agents
- Subject: Cursor

Cursor introduces new tools for teams to configure development environments for cloud agents, including multi-repo support, Dockerfile-based configuration with build secrets and layer caching, improved agent-led setup, and environment governance with version history and audit logs.

### Cursor — Full-screen Tabs and Compact Chats

- Company: Cursor (cursor.com)
- Announced: 2026-05-13
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/3-4
- Record: https://forck.live/items/2608-full-screen-tabs-and-compact-chats
- Subject: Cursor

Cursor introduces full-screen tabs for the Agents Window, compact chat responses with customizable tool call density, and various improvements and bug fixes.

### OpenAI — Our response to the TanStack npm supply chain attack

- Company: OpenAI (openai.com)
- Announced: 2026-05-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/our-response-to-the-tanstack-npm-supply-chain-attack
- Record: https://forck.live/items/129-our-response-to-the-tanstack-npm-supply-chain-attack
- Subject: GPT / ChatGPT / API

OpenAI published a response to the TanStack npm supply chain attack, describing the protections it implemented and the requirement for macOS users to update OpenAI apps by June 12, 2026.

### OpenAI — Finance workflows with ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-05-12T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/how-finance-teams-use-codex
- Record: https://forck.live/items/4147-finance-workflows-with-chatgpt-work
- Subject: GPT / ChatGPT / API

Finance teams can use ChatGPT Work to build various financial documents and analyses from real work inputs.

### OpenAI — How finance teams use ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-05-12T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/how-finance-teams-use-codex
- Record: https://forck.live/items/130-how-finance-teams-use-chatgpt-work
- Subject: GPT / ChatGPT / API

Finance teams can use ChatGPT Work to build various financial reports and scenarios, but it does not announce any new model, product launch, or other specific category. It is a how-to guide or example use case.

### Google DeepMind — Co-Scientist: A multi-agent AI partner to accelerate research

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-12T14:40:07+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/
- Record: https://forck.live/items/1234-co-scientist-a-multi-agent-ai-partner-to-accelerate-research
- Subject: Gemini

Google DeepMind announces Co-Scientist, a collaborative AI research partner built on Gemini, aimed at accelerating scientific breakthroughs.

### Vercel — AI Gateway production index

- Company: Vercel (vercel.com)
- Announced: 2026-05-12T04:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/ai-gateway-production-index
- Record: https://forck.live/items/20709-ai-gateway-production-index
- Subject: v0 / AI SDK / AI Gateway

Ask which AI model is best, and the answer changes before the ink dries. That's what happens in an industry where new models are released weekly.

### Isomorphic Labs — Isomorphic Labs announces Series B investment round

- Company: Isomorphic Labs (isomorphiclabs.com)
- Announced: 2026-05-12
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.isomorphiclabs.com/articles/isomorphic-labs-announces-series-b-investment-round
- Record: https://forck.live/items/19133-isomorphic-labs-announces-series-b-investment-round
- Subject: Drug Design Engine / AlphaFold

Isomorphic Labs raised $2.1 billion in Series B funding, led by Thrive Capital, with participation from Alphabet, GV, MGX, Temasek, CapitalG, and the UK Sovereign AI Fund. The company plans to use the capital to advance its AI drug design engine (IsoDDE) across multiple therapeutic areas and drug modalities.

### Perceptron — Introducing Perceptron Mk1

- Company: Perceptron (perceptron.inc)
- Announced: 2026-05-12
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.perceptron.inc/blog/introducing-perceptron-mk1
- Record: https://forck.live/items/18615-introducing-perceptron-mk1
- Subject: Perceptron Isaac / Perceptron Mk1

Perceptron released Mk1, a closed-source model for video understanding and embodied reasoning. It matches frontier models like Gemini Pro on video and embodied reasoning while costing less than Gemini Flash Lite. Mk1 features temporal reasoning with thinking traces, temporal grounding with video clips, in-context learning, and advanced image reasoning including pointing, counting, OCR, and structured document extraction.

### Odyssey — Introducing PROWL-1: Learning Through Discovery

- Company: Odyssey (odyssey.systems)
- Announced: 2026-05-12
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://odyssey.systems/introducing-prowl-1
- Record: https://forck.live/items/18597-introducing-prowl-1-learning-through-discovery
- Subject: Odyssey world models

Odyssey introduced PROWL-1, a reinforcement learning-driven adversarial framework that uses an RL agent to explore game environments and discover failures in world models, which are then used to improve the model through a closed-loop training cycle. The framework was evaluated in Minecraft, showing improvements in action-following, visual quality, temporal coherence, and out-of-distribution robustness compared to a baseline.

### Inception — Mercury 2 and the Rise of Real-time Subagents

- Company: Inception (inceptionlabs.ai)
- Announced: 2026-05-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.inceptionlabs.ai/blog/rise-of-realtime-subagents
- Record: https://forck.live/items/18521-mercury-2-and-the-rise-of-real-time-subagents
- Subject: Mercury

Inception's Mercury 2 diffusion LLM is used by Augment Code as a dedicated subagent for context compaction, model routing, and tool search in production coding agents. Augment Code reports that switching to Mercury 2 reduced context compaction latency by 82% and cost by 90% while maintaining quality, and that Mercury 2 returns tool summaries in under a second. Inception's evaluation shows Mercury 2 is roughly 5x faster than Sonnet 4.6 with matching quality on compaction tasks.

### Genspark — How One CIO Built 25+ Internal Tools Without Developers

- Company: Genspark (genspark.ai)
- Announced: 2026-05-12
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.genspark.ai/blog/ficofi-cio-ai-internal-tools
- Record: https://forck.live/items/18255-how-one-cio-built-25-internal-tools-without-developers
- Subject: Genspark

### Krea — Krea 2 Large shines in style fidelity

- Company: Krea (krea.ai)
- Announced: 2026-05-12
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/krea-2-style-transfer-benchmark
- Record: https://forck.live/items/15082-krea-2-large-shines-in-style-fidelity
- Subject: Krea 2 / AI creative platform

Contra Labs evaluated Krea 2 Large against GPT Image 2, Gemini 3 Pro Image Preview, and Seedream 5.0 Lite on style-transfer tasks. Krea 2 Large ranked second in Style Fidelity, 0.14 points behind GPT Image 2, and was the only non-GPT model to score above the neutral midpoint on that dimension. At the top of the scale, Krea 2 Large received a 5 on Style Fidelity for 23% of outputs, matching GPT Image 2's 22% rate.

### Krea — What Is Krea 2? Krea's Foundation Image Model

- Company: Krea (krea.ai)
- Announced: 2026-05-12
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.krea.ai/blog/krea-2-image-model
- Record: https://forck.live/items/15081-what-is-krea-2-krea-s-foundation-image-model
- Subject: Krea 2 / AI creative platform

Krea released Krea 2, its first foundation image model built from scratch, focused on aesthetics, style transfer, and creative control. The model includes a style transfer system that lets users pass in reference images, control style influence, combine multiple styles, and adjust variation across a batch. Krea also announced a Turbo variant for faster iteration.

### Together AI — Introducing voice finder — a new tool to quickly find the right voice for your app from over 600+ voices

- Company: Together AI (together.ai)
- Announced: 2026-05-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/introducing-voice-finder-a-new-tool-to-quickly-find-the-right-voice-for-your-app-from-over-600-voices
- Record: https://forck.live/items/2348-introducing-voice-finder-a-new-tool-to-quickly-find-the-right-voice-for-your
- Subject: Inference platform

Together AI launched Voice finder, a tool that indexes over 600 voices across 10 TTS models, allowing developers to search by text description or audio sample, compare recommendations, and filter by attributes like pitch, accent, and emotion.

### OpenAI — AutoScout24 scales engineering with AI-powered workflows

- Company: OpenAI (openai.com)
- Announced: 2026-05-12
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/autoscout24
- Record: https://forck.live/items/133-autoscout24-scales-engineering-with-ai-powered-workflows
- Subject: GPT / ChatGPT / API

AutoScout24 Group uses Codex and ChatGPT to speed development cycles, improve code quality, and expand AI adoption.

### OpenAI — How NVIDIA engineers and researchers build with Codex

- Company: OpenAI (openai.com)
- Announced: 2026-05-12
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/nvidia
- Record: https://forck.live/items/132-how-nvidia-engineers-and-researchers-build-with-codex
- Subject: GPT / ChatGPT / API

NVIDIA engineers and researchers use Codex with GPT-5.5 to build production systems and convert research ideas into experiments.

### OpenAI — What Parameter Golf taught us about AI-assisted research

- Company: OpenAI (openai.com)
- Announced: 2026-05-12
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/what-parameter-golf-taught-us
- Record: https://forck.live/items/131-what-parameter-golf-taught-us-about-ai-assisted-research
- Subject: GPT / ChatGPT / API

OpenAI describes Parameter Golf, an event with over 1,000 participants and 2,000 submissions, focused on AI-assisted machine learning research, coding agents, quantization, and novel model design under strict constraints.

### Cursor — Cursor in Microsoft Teams

- Company: Cursor (cursor.com)
- Announced: 2026-05-11T23:59:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/microsoft-teams
- Record: https://forck.live/items/2610-cursor-in-microsoft-teams
- Subject: Cursor

Cursor is now available as an integration in Microsoft Teams, allowing users to mention @Cursor to delegate tasks or pull information.

### Amazon — Building Blocks for Foundation Model Training and Inference on AWS

- Company: Amazon (amazon.com)
- Announced: 2026-05-11T23:18:26+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/amazon/foundation-model-building-blocks
- Record: https://forck.live/items/1505-building-blocks-for-foundation-model-training-and-inference-on-aws

The infrastructure requirements for foundation model training and inference on AWS, focusing on compute, network, and storage building blocks, and how they integrate with open-source software stacks across the model lifecycle.

### Harvey — Harvey Opens in Paris

- Company: Harvey (harvey.ai)
- Announced: 2026-05-11T16:16:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/harvey-opens-in-paris
- Record: https://forck.live/items/17568-harvey-opens-in-paris
- Subject: Harvey

Harvey announced the opening of its Paris office as part of its EMEA expansion, which already includes Dublin, London, Madrid, and Munich. The company also stated that the French law firm August Debouzy is adopting Harvey firmwide. Harvey reiterated that it never trains on customer data and requires zero data retention from all model providers.

### OpenAI — How ChatGPT adoption broadened in early 2026

- Company: OpenAI (openai.com)
- Announced: 2026-05-11T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/signals/research/2026q1-update
- Record: https://forck.live/items/134-how-chatgpt-adoption-broadened-in-early-2026
- Subject: GPT / ChatGPT / API

ChatGPT's user base broadened in Q1 2026, with notable increases among users over 35 and a more balanced gender distribution, indicating wider mainstream acceptance.

### Runway — How Real-Time Video Generation Is Changing Online Interaction

- Company: Runway (runway.com)
- Announced: 2026-05-11T10:10:10.110000+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/research/how-real-time-video-generation-is-changing-online-interaction
- Record: https://forck.live/items/7883-how-real-time-video-generation-is-changing-online-interaction
- Subject: Gen / Aleph

Runway announced Runway Characters, a real-time video generation product built on GWM-1 that produces expressive conversational characters from a single reference image. The product is available now for developers via API and in the Runway web app. The post also outlines the broader vision for real-time video generation across gaming, education, training, and customer experience.

### OpenAI — How enterprises are scaling AI

- Company: OpenAI (openai.com)
- Announced: 2026-05-11T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/business/guides-and-resources/how-enterprises-are-scaling-ai
- Record: https://forck.live/items/136-how-enterprises-are-scaling-ai
- Subject: GPT / ChatGPT / API

OpenAI published a guide titled 'How enterprises scale AI: from early experiments to compounding impact through trust, governance, workflow design, and quality at scale.'

### OpenAI — OpenAI Campus Network: Student club interest form

- Company: OpenAI (openai.com)
- Announced: 2026-05-11T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/openai-campus-network-student-club-interest-form
- Record: https://forck.live/items/135-openai-campus-network-student-club-interest-form
- Subject: GPT / ChatGPT / API

OpenAI invites students to join the Campus Network to connect clubs, access AI tools, host events, and build an AI-powered campus community.

### OpenAI — OpenAI launches DeployCo to help businesses build around intelligence

- Company: OpenAI (openai.com)
- Announced: 2026-05-11T06:00:00+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/openai-launches-the-deployment-company
- Record: https://forck.live/items/137-openai-launches-deployco-to-help-businesses-build-around-intelligence
- Subject: GPT / ChatGPT / API

OpenAI announced the launch of DeployCo, a new enterprise deployment company aimed at helping organizations integrate frontier AI into production and achieve measurable business impact.

### Thinking Machines Lab — Interaction Models: A Scalable Approach to Human-AI Collaboration

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2026-05-11
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://thinkingmachines.ai/blog/interaction-models/
- Record: https://forck.live/items/18567-interaction-models-a-scalable-approach-to-human-ai-collaboration
- Subject: Thinking Machines / Inkling

Thinking Machines Lab announced a research preview of interaction models, which are trained from scratch to handle audio, video, and text natively in real time using a multi-stream, micro-turn design. The preview demonstrates new interaction capabilities and state-of-the-art combined performance in intelligence and responsiveness.

### Runway — Stop Writing YAML: Configuring ML Systems with confingy

- Company: Runway (runway.com)
- Announced: 2026-05-11
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/engineering/stop-writing-yaml-configuring-ml-systems-with-confingy
- Record: https://forck.live/items/7884-stop-writing-yaml-configuring-ml-systems-with-confingy
- Subject: Gen / Aleph

Runway published confingy, an open source library for configuring Python-based ML systems. The post discusses the pitfalls of YAML configuration and introduces confingy as a solution.

### Cognition — How to Automate Failure Triages and 10x Test Generation: What We've Learned Deploying AI Across HIL/SIL Workflows

- Company: Cognition (cognition.com)
- Announced: 2026-05-11
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cognition.com/blog/how-to-automate-failure-triages-and-10x-test-generation-what-weve-learned-deploying-ai-across-hilsil-workflows
- Record: https://forck.live/items/7662-how-to-automate-failure-triages-and-10x-test-generation-what-we-ve-learned
- Subject: Devin / SWE

Cognition describes how its AI agent Devin has been deployed across HIL/SIL workflows at OEMs, including RV Tech and Mercedes, to automate failure triaging, speed up test generation, and convert HIL tests to SIL equivalents. The post reports outcomes such as reclaiming 2,000–4,000 engineering hours per month for one team, a 10x increase in test-generation at RV Tech, and reducing program test development from half a week to one day for a large US automotive company.

### Cursor — Bugbot Effort Levels

- Company: Cursor (cursor.com)
- Announced: 2026-05-11
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/05-11-26
- Record: https://forck.live/items/2611-bugbot-effort-levels
- Subject: Cursor

Cursor now allows team admins and Individual plan users to customize the effort level for Bugbot's PR reviews, offering Default, High, and Custom configurations. High effort increases bug detection but costs more and takes longer. Custom effort lets users describe when to use default or high effort. The feature requires usage-based billing.

### Together AI — Serving DeepSeek-V4: why million-token context is an inference systems problem

- Company: Together AI (together.ai)
- Announced: 2026-05-11
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.together.ai/blog/serving-deepseek-v4-why-million-token-context-is-an-inference-systems-problem
- Record: https://forck.live/items/2349-serving-deepseek-v4-why-million-token-context-is-an-inference-systems-problem
- Subject: Inference platform

Together AI discusses serving implications of DeepSeek-V4, focusing on its 1M-token context window and architectural changes with compressed sparse attention mechanisms. They describe how V4's token-axis KV cache compression improves serving capacity but requires sophisticated memory management across multiple cache types (CSA, HCA, SWA). The post shares early bring-up results on NVIDIA HGX B200 hardware, showing increased KV-cache capacity from 1.2M to 3.7M tokens on a single node through cache policy optimizations.