# forck.live — 29 June – 5 July 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-W27, 29 June – 5 July 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-W27
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Sakana AI — Learning Multi-Agent Coordination via Sheaf-ADMM

- Company: Sakana AI (sakana.ai)
- Announced: 2026-07-04T15:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/sheaf-admm
- Record: https://forck.live/items/4679-learning-multi-agent-coordination-via-sheaf-admm
- Subject: Sakana models

Sakana AI announced a research paper on a multi-agent coordination framework called Sheaf-ADMM, to be presented at ICML 2026.

### Sakana AI — Bridging Spherical Black-Box Optimizers

- Company: Sakana AI (sakana.ai)
- Announced: 2026-07-03T15:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/bbob
- Record: https://forck.live/items/4680-bridging-spherical-black-box-optimizers
- Subject: Sakana models

Sakana AI announces a research paper at ICML 2026 that bridges parametric and nonparametric black-box optimization methods, showing they are variations of a single update equation. The paper introduces two new hybrid optimizers, AdaPol and SchedPol, and demonstrates their application in merging large language models efficiently.

### Lovable — $85,000 in tokens later: What I learned from scaling agentic coding at Lovable

- Company: Lovable (lovable.dev)
- Announced: 2026-07-03T09:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/85000-in-tokens-later-scaling-agentic-coding-at-lovable
- Record: https://forck.live/items/12071-85-000-in-tokens-later-what-i-learned-from-scaling-agentic-coding-at-lovable
- Subject: Lovable / AI app builder

A Lovable engineer describes scaling agentic coding from 20-30 merged PRs per week in January to over 150 per week by June, using a hierarchy of agents and automated review. The post details a shift from human line-by-line code review to AI-driven review, with human attention reserved for high-impact design decisions. It also covers a risk classification system for PRs that routes changes to fast AI, slow AI, or human review based on risk level.

### Amp — More Orb Sizes

- Company: Amp (ampcode.com)
- Announced: 2026-07-03
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/more-orb-sizes
- Record: https://forck.live/items/7732-more-orb-sizes
- Subject: Amp

Amp now allows users to select the size of the orbs used to run agents remotely. The storage per orb has also been doubled from 20GB to 40GB at no extra cost.

### Microsoft — Meet Brain: The AI system behind Azure reliability

- Company: Microsoft (microsoft.com)
- Announced: 2026-07-02T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://azure.microsoft.com/en-us/blog/meet-brain-the-ai-system-behind-azure-reliability/
- Record: https://forck.live/items/2451-meet-brain-the-ai-system-behind-azure-reliability
- Subject: Azure AI

Microsoft introduces Brain, an AI-powered cloud reliability intelligence system for Azure that integrates platform telemetry, AI/ML models, service dependencies, and customer impact to provide a continuously updated view of service, region, and workload health, and already powers customer health notifications, deployment safeguards, and outage declaration.

### Ai2 — Modular LLMs at scale: how FlexOlmo is helping to pool national expertise without pooling sensitive data

- Company: Ai2 (allenai.org)
- Announced: 2026-07-02
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://allenai.org/blog/flexmore
- Record: https://forck.live/items/18547-modular-llms-at-scale-how-flexolmo-is-helping-to-pool-national-expertise
- Subject: Olmo / Molmo / Asta

Researchers at the University of Southern Denmark's OdenseNLP lab built FlexMoRE, a modification of Ai2's FlexOlmo architecture that replaces most full-size experts with smaller low-rank adapters. FlexMoRE outperforms a FlexOlmo-style baseline while using less than one-third the parameters, enabling modular language models to run on consumer hardware. The work is part of the Danish Foundation Models (DFM) project, which aims to develop open Danish-language models for institutions that cannot share sensitive data.

### Runway — Borrowing the Night: Reclaiming Idle Inference GPUs for Research

- Company: Runway (runway.com)
- Announced: 2026-07-02
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://runway.com/news/engineering/borrowing-the-night-reclaiming-idle-inference-gpus-for-research
- Record: https://forck.live/items/7869-borrowing-the-night-reclaiming-idle-inference-gpus-for-research
- Subject: Gen / Aleph

Runway built a capacity controller called deckard that reallocates GPUs between production inference and research based on daily demand cycles, using queueing theory to optimize allocations so that production tracks demand without over-provisioning and more GPUs are available for research overnight.

### Amp — Read Bigger Threads

- Company: Amp (ampcode.com)
- Announced: 2026-07-02
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/read-bigger-threads
- Record: https://forck.live/items/7733-read-bigger-threads
- Subject: Amp

Amp rewrote its read_thread tool, which lets the platform pull context from other threads, to handle long, compacted threads. The new version uses a subagent that searches and extracts information from long threads, switching the underlying model to GLM 5.2 from Gemini 3.5 Flash and optimizing for correctness over speed. The change requires no user action.

### Cognition — Introducing the Devin Security Vulnerability Remediation Program

- Company: Cognition (cognition.com)
- Announced: 2026-07-02
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-security-vulnerability-remediation-program
- Record: https://forck.live/items/7650-introducing-the-devin-security-vulnerability-remediation-program
- Subject: Devin / SWE

Cognition announced the Devin Security Vulnerability Remediation Program, where its forward-deployed engineering team embeds with clients and uses the Devin AI software engineer to find, validate, and fix vulnerabilities. The program has two phases: clearing the existing vulnerability backlog and establishing continuous discovery and remediation via Devin Security Swarm. It integrates with existing scanners such as Snyk, SonarQube, and Checkmarx, and is available to enterprise customers deploying Devin Cloud at meaningful scale.

### Mistral AI — Leanstral 1.5: Proof Abundance for All

- Company: Mistral AI (mistral.ai)
- Announced: 2026-07-02
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/leanstral-1-5
- Record: https://forck.live/items/3290-leanstral-1-5-proof-abundance-for-all
- Subject: Mistral / Le Chat

Mistral AI announces Leanstral 1.5, a research model focused on proof abundance.

### Databricks — How we keep GPUs reliable across Databricks AI

- Company: Databricks (databricks.com)
- Announced: 2026-07-01T23:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.databricks.com/blog/how-we-keep-gpus-reliable-across-databricks-ai
- Record: https://forck.live/items/17508-how-we-keep-gpus-reliable-across-databricks-ai
- Subject: Mosaic AI

Databricks AI describes the three main categories of GPU failures at scale—crashed jobs, silent slowdowns, and numerical corruption—and outlines its approach to maintaining reliability through stress testing with diverse workloads and a multi-stage health check system that validates hardware before workloads start, monitors for degradation under load, and probes inter-node fabric health.

### Google DeepMind — The latest AI news we announced in June 2026

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-07-01T18:15:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-june-2026/
- Record: https://forck.live/items/1304-the-latest-ai-news-we-announced-in-june-2026
- Subject: Gemini

Google has published a summary of AI updates from June 2026, but provides no specific details about any announcements.

### Meta — Meta’s AI Storage Blueprint at Scale

- Company: Meta (meta.com)
- Announced: 2026-07-01T16:00:36+00:00
- Category: infrastructure-release
- Coverage: 1 outlet
- Announcement: yes
- Group: covered
- Source: https://engineering.fb.com/2026/07/01/data-infrastructure/metas-ai-storage-blueprint-at-scale/
- Record: https://forck.live/items/1423-meta-s-ai-storage-blueprint-at-scale
- Subject: Llama / infrastructure

Meta describes the evolution of its BLOB-storage architecture to support AI workloads, focusing on maximizing GPU utilization and research velocity, with a migration from legacy HDD-based to flash-based storage and a simplified metadata path to reduce latency and stalls.

### Microsoft — Proving application resilience on Azure with Chaos Studio

- Company: Microsoft (microsoft.com)
- Announced: 2026-07-01T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://azure.microsoft.com/en-us/blog/proving-application-resilience-on-azure-with-chaos-studio/
- Record: https://forck.live/items/2452-proving-application-resilience-on-azure-with-chaos-studio
- Subject: Azure AI

Microsoft announces the public preview of Azure Chaos Studio Workspaces, a scenario-focused feature that enables resilience testing using curated outage scenarios like Zone Down, DNS Outage, and SQL failover, simplifying chaos engineering for Azure customers.

### Lovable — How Nursa built a new product in 48 hours and changed how its entire company ships software

- Company: Lovable (lovable.dev)
- Announced: 2026-07-01T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/how-nursa-built-a-new-product-in-48-hours-and-changed-how-its-entire-company-ships-software
- Record: https://forck.live/items/12072-how-nursa-built-a-new-product-in-48-hours-and-changed-how-its-entire-company
- Subject: Lovable / AI app builder

A new business line, built in a weekend. Nursa Study, a net-new product for nursing students and colleges, went from prototype to product in a single weekend, and now has paying enterprise customers.

### Runway — Runway Announces Creative Partnership with Bertelsmann

- Company: Runway (runway.com)
- Announced: 2026-07-01T10:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/runway-announces-creative-partnership-with-bertelsmann
- Record: https://forck.live/items/7870-runway-announces-creative-partnership-with-bertelsmann
- Subject: Gen / Aleph

Runway announces a creative partnership with Bertelsmann, integrating Runway's AI models across Bertelsmann's global portfolio of businesses including RTL Group, BMG, and Bertelsmann Marketing Services. The partnership builds on a prior collaboration that began in 2024.

### Kyutai — Surflo: Consistent 3D surfaces from a global state

- Company: Kyutai (kyutai.org)
- Announced: 2026-07-01
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://kyutai.org/blog/2026-07-01-surflo
- Record: https://forck.live/items/18338-surflo-consistent-3d-surfaces-from-a-global-state
- Subject: Moshi / Pocket TTS

Kyutai's research introduces Surflo, a method that maps any number of views into consistent 3D surfaces at arbitrary resolution via a global state.

### Factory AI — Droid Shield 2.0: learned secret detection

- Company: Factory AI (factory.com)
- Announced: 2026-07-01
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/droid-shield-2-0
- Record: https://forck.live/items/17829-droid-shield-2-0-learned-secret-detection
- Subject: Droid

Factory's Droids write, refactor, and commit code autonomously at a volume that no human reviewer can completely monitor. Keeping autonomous software engineering safe at enterprise scale means secret detection that is trustworthy but doesn't introduce unnecessary friction.

### Cognition — Introducing Devin Security Swarm

- Company: Cognition (cognition.com)
- Announced: 2026-07-01
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cognition.com/blog/introducing-devin-security-swarm
- Record: https://forck.live/items/7651-introducing-devin-security-swarm
- Subject: Devin / SWE

Cognition introduced Devin Security Swarm, a product that finds, validates, and fixes code vulnerabilities at scale. It uses parallel agents to reason across codebases, confirm exploitability at runtime, and ship remediation PRs. The product is available starting today and is already used by major companies.

### xAI — Introducing the Voice Agent Builder

- Company: xAI (x.ai)
- Announced: 2026-07-01
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.ai/news/grok-voice-agent-builder
- Record: https://forck.live/items/3445-introducing-the-voice-agent-builder
- Subject: Grok

xAI introduces the Voice Agent Builder, a no-code tool that lets users create a personalized voice agent in under 2 minutes.

### Together AI — Announcing our $800M Series C to accelerate the shift to open-source AI

- Company: Together AI (together.ai)
- Announced: 2026-07-01
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/announcing-our-series-c
- Record: https://forck.live/items/2338-announcing-our-800m-series-c-to-accelerate-the-shift-to-open-source-ai
- Subject: Inference platform

Together AI announced an $800 million Series C funding round from investors including Aramco Ventures, NVIDIA, Vista Equity, General Catalyst, and others, with additional commitments for over 500 MW of compute capacity to support growth in open-source AI inference and training infrastructure.

### Hugging Face — Hugging Face and Cerebras bring Gemma 4 to real-time voice AI

- Company: Hugging Face (huggingface.co)
- Announced: 2026-07-01
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/cerebras-gemma4-voice-ai
- Record: https://forck.live/items/1467-hugging-face-and-cerebras-bring-gemma-4-to-real-time-voice-ai
- Subject: Platform

Hugging Face and Cerebras announce a real-time speech-to-speech pipeline using Google DeepMind's Gemma 4 VLM on Cerebras hardware for low-latency inference, with an open, modular architecture that includes Nvidia's Parakeet for speech recognition and Alibaba's Qwen3TTS for text-to-speech. The system is demonstrated in a demo and the code is available on GitHub. The collaboration aims to improve latency and naturalness in voice AI interactions.

### Hugging Face — ScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration

- Company: Hugging Face (huggingface.co)
- Announced: 2026-06-30T18:32:50+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/ibm-research/scarfbench
- Record: https://forck.live/items/1468-scarfbench-benchmarking-ai-agents-for-enterprise-java-framework-migration
- Subject: Platform

IBM Research introduces ScarfBench, an open benchmark for evaluating AI agents on enterprise Java framework migration tasks. The benchmark includes 34 applications, 102 framework implementations, 204 migration tasks, about 151K lines of code, and 1,331 expert-written tests. Evaluations show that even the strongest agents achieve less than 10% behavioral success, agents are overconfident in reporting build success (Claude Code reported 29/30 builds successful but only 22 actually built), migration is iterative, configuration dominates effort, and environment and tooling matter. The leaderboard is available at scarfbench.info/leaderboard.

### Harvey — Claude Sonnet 5, Now Live in Harvey

- Company: Harvey (harvey.ai)
- Announced: 2026-06-30T18:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/sonnet-5-in-harvey
- Record: https://forck.live/items/17545-claude-sonnet-5-now-live-in-harvey
- Subject: Harvey

Harvey is making Anthropic's Claude Sonnet 5 available to its customers. The model achieved a 5.8% all-pass rate on Harvey's Legal Agent Benchmark and a 91.3% score on BigLaw Bench, the highest recorded across Sonnet and Opus models. Availability begins with eligible US customers in the coming days, with EU and AU availability to be announced.

### Google Research — Expanding our Heat Resilience data to 50+ global cities

- Company: Google Research (research.google)
- Announced: 2026-06-30T17:03:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/expanding-our-heat-resilience-data-to-50-global-cities/
- Record: https://forck.live/items/1326-expanding-our-heat-resilience-data-to-50-global-cities
- Subject: Research

Google Research expands building-level rooftop reflectivity dataset to over 50 global cities, and launches a new Heat Resilience Earth Engine App to make the data accessible.

### Microsoft Research — SkillOpt: Agent skills as trainable parameters

- Company: Microsoft Research (microsoft.com)
- Announced: 2026-06-30T16:50:02+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.microsoft.com/en-us/research/blog/skillopt-agent-skills-as-trainable-parameters/
- Record: https://forck.live/items/2461-skillopt-agent-skills-as-trainable-parameters
- Subject: Research / Phi

SkillOpt is a method that treats agent skill files as trainable parameters outside a frozen model, using a training-style optimization loop to improve agent performance without changing model weights. It achieves consistent gains across 52 evaluation cells spanning six benchmarks, seven target models, and three execution modes, outperforming human-written skills and other baseline methods. The optimized skills remain compact, auditable, and transferable across model scales and tasks.

### Meta — 10 Years of Meta’s Commitment to Python

- Company: Meta (meta.com)
- Announced: 2026-06-30T16:00:46+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://engineering.fb.com/2026/06/30/open-source/10-years-of-metas-commitment-to-python/
- Record: https://forck.live/items/1424-10-years-of-meta-s-commitment-to-python
- Subject: Llama / infrastructure

Meta is celebrating 10 years as a sponsor of the Python Software Foundation.

### Vercel — Vercel Ship 2026 recap

- Company: Vercel (vercel.com)
- Announced: 2026-06-30T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/vercel-ship-2026-recap
- Record: https://forck.live/items/20690-vercel-ship-2026-recap
- Subject: v0 / AI SDK / AI Gateway

For a decade, Vercel has shaped how the web is built. Now, we’re doing the same for agents.

### Vercel — Run any Dockerfile on Vercel

- Company: Vercel (vercel.com)
- Announced: 2026-06-30T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/dockerfile-on-vercel
- Record: https://forck.live/items/20689-run-any-dockerfile-on-vercel
- Subject: v0 / AI SDK / AI Gateway

Vercel announced that any Dockerfile can now be deployed on its platform. Adding a Dockerfile.vercel file to a project causes Vercel to build, store, deploy, and autoscale the image on Fluid compute, with billing based on active CPU time rather than wall time. The feature supports any language or framework that listens on $PORT and includes preview deployments, autoscaling, and integrated observability.

### Microsoft — Azure IaaS: How to design, build, and optimize cloud infrastructure for long-term cost efficiency

- Company: Microsoft (microsoft.com)
- Announced: 2026-06-30T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://azure.microsoft.com/en-us/blog/azure-iaas-how-to-design-build-and-optimize-cloud-infrastructure-for-long-term-cost-optimization-efficiency/
- Record: https://forck.live/items/2453-azure-iaas-how-to-design-build-and-optimize-cloud-infrastructure-for-long-term
- Subject: Azure AI

Best practices and guidance for optimizing cost efficiency in Azure IaaS environments across compute, storage, and networking. It discusses common inefficiencies like overprovisioning and misaligned storage tiers, and recommends Azure services such as VM Scale Sets, Compute Fleet, flexible pricing options, and automated data lifecycle management to reduce total cost of ownership.

### Hugging Face — Why Specialization Is Inevitable

- Company: Hugging Face (huggingface.co)
- Announced: 2026-06-30T14:39:11+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/Dharma-AI/why-specialization-is-inevitable
- Record: https://forck.live/items/1469-why-specialization-is-inevitable
- Subject: Platform

The inevitability of specialization in AI, drawing on optimization theory, evolutionary biology, competitive markets, and machine learning, and interprets ideas from the 2026 paper 'AI Must Embrace Specialization via Superhuman Adaptable Intelligence' by Goldfeder, Wyder, LeCun, and Shwartz-Ziv.

### ElevenLabs — Introducing Procedures in ElevenAgents

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-06-30T12:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://elevenlabs.io/blog/procedures
- Record: https://forck.live/items/3346-introducing-procedures-in-elevenagents
- Subject: Voice AI

ElevenLabs introduced Procedures in their ElevenAgents product.

### Google Research — Introducing TabFM: A zero-shot foundation model for tabular data

- Company: Google Research (research.google)
- Announced: 2026-06-30T10:26:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://research.google/blog/introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data/
- Record: https://forck.live/items/1327-introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data
- Subject: Research

Google Research introduces TabFM, a zero-shot foundation model for tabular data that performs classification and regression without traditional training. The model is available on Hugging Face and GitHub under the TabFM name.

### OpenAI — How ChatGPT adoption has expanded

- Company: OpenAI (openai.com)
- Announced: 2026-06-30T09:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/how-chatgpt-adoption-has-expanded
- Record: https://forck.live/items/45-how-chatgpt-adoption-has-expanded
- Subject: GPT / ChatGPT / API

OpenAI signals data shows global growth in ChatGPT adoption, with increased usage and exploration of capabilities across regions and languages.

### Vercel — Vercel Services: Run full stack on Vercel

- Company: Vercel (vercel.com)
- Announced: 2026-06-30T04:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/vercel-services-run-full-stack-on-vercel
- Record: https://forck.live/items/20692-vercel-services-run-full-stack-on-vercel
- Subject: v0 / AI SDK / AI Gateway

Vercel introduced Vercel Services in public beta, allowing multiple frameworks (e.g., Next.js frontend, FastAPI backend) to run in one Vercel Project with atomic deployments, shared preview deployments, and internal service communication without routing through the public internet. The feature is configured via the services key in vercel.json and supports service bindings for internal traffic, auto-detection of frameworks, and integration with Vercel's platform including Fluid compute, WebSocket support, Vercel Connect for external services, and database provisioning from the Marketplace.

### Vercel — Vercel and Shopify are rebuilding Hydrogen

- Company: Vercel (vercel.com)
- Announced: 2026-06-30T04:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/vercel-and-shopify-are-rebuilding-hydrogen
- Record: https://forck.live/items/20691-vercel-and-shopify-are-rebuilding-hydrogen
- Subject: v0 / AI SDK / AI Gateway

Vercel and Shopify announced a rebuild of Hydrogen, the headless storefront framework, making it open source and runtime agnostic. The new version includes three layers: core (shared JavaScript for the Shopify API), client (handles cart state management), and server (full-stack framework integration with guidance and templates).

### Supabase — Agentic Coding on Supabase with OpenCode

- Company: Supabase (supabase.com)
- Announced: 2026-06-30
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/agentic-coding-on-supabase-with-opencode
- Record: https://forck.live/items/21282-agentic-coding-on-supabase-with-opencode
- Subject: Supabase

Supabase announced an integration with OpenCode, an open-source AI coding agent. The integration allows agents to manage Supabase accounts and projects via APIs and bundled skills, and guides users through connecting Supabase MCP for database, Edge Functions, and logs. It requires OpenCode 1.3.4 or later and is open source.

### Thinking Machines Lab — Learning to Replicate Expert Judgment in Financial Tasks

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2026-06-30
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks
- Record: https://forck.live/items/18557-learning-to-replicate-expert-judgment-in-financial-tasks
- Subject: Thinking Machines / Inkling

Thinking Machines Lab trained a proprietary model that outperforms frontier models on six financial information filtering tasks, achieving higher accuracy and recall at lower cost. The model was fine-tuned on high-quality human annotations to replicate expert judgment in tasks such as document relevancy classification and boilerplate identification.

### Liquid AI — IFStruct: Measuring structured-output compliance

- Company: Liquid AI (liquid.ai)
- Announced: 2026-06-30
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.liquid.ai/blog/ifstruct-v1.0
- Record: https://forck.live/items/16780-ifstruct-measuring-structured-output-compliance
- Subject: LFM / d1 models

Liquid AI introduced IFStruct, a generative benchmark for evaluating language models' ability to produce valid structured output and follow diverse schema requirements. The benchmark tests output validity and schema following, with tasks calibrated so frontier models can achieve near-100% accuracy. Small models like LFM2.5-350M, trained with RL on a dedicated held-out training split, can exceed the performance of far larger models like Qwen3.5-4B and granite-4.0-h-tiny.

### Manus — Mobbin connector on Manus: Turn Proven App Designs Into Finished Work

- Company: Manus (manus.im)
- Announced: 2026-06-30
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://manus.im/blog/manus-mobbin-connector
- Record: https://forck.live/items/14709-mobbin-connector-on-manus-turn-proven-app-designs-into-finished-work
- Subject: Manus

Manus has released a connector for the Mobbin design research library, allowing users to search screens, flows, and website sections within Manus and then build apps, slide decks, analyses, or critiques directly from that research. The connector is read-only and requires a paid Mobbin plan.

### Suno — How Dream Relic sees sound and gets it stuck in your head

- Company: Suno (suno.com)
- Announced: 2026-06-30
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://suno.com/blog/dream-relic
- Record: https://forck.live/items/10319-how-dream-relic-sees-sound-and-gets-it-stuck-in-your-head
- Subject: Music / Studio

Suno profiles AI visual artist Dream Relic, who uses Suno's Create feature to turn his old lyrics into songs for his audiovisual project. He describes how the tool helped him bridge the gap between his musical vision and technical ability, and notes that a viral video response led him to take the music side of his work more seriously.

### Cartesia — A Beginner's Guide to Voice AI Terminology

- Company: Cartesia (cartesia.ai)
- Announced: 2026-06-30
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.cartesia.ai/blog/voice-terminology
- Record: https://forck.live/items/8011-a-beginner-s-guide-to-voice-ai-terminology
- Subject: Sonic / Ink

Cartesia published a guide to voice AI terminology, explaining concepts such as STT, VAD, TTS, and turn detection, and referencing its own models Ink, Ink-2, and Sonic.

### Cartesia — A Guide to Choosing Voice AI Models

- Company: Cartesia (cartesia.ai)
- Announced: 2026-06-30
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.cartesia.ai/blog/choosing-voice-ai-models
- Record: https://forck.live/items/8010-a-guide-to-choosing-voice-ai-models
- Subject: Sonic / Ink

Cartesia published a guide on selecting voice AI models for enterprise voice agents, emphasizing that real-world performance depends on use-case-specific conditions such as telephony infrastructure, noise, and conversational dynamics. The guide recommends evaluating models on metrics like Time-to-Complete-Transcript (TTCT) rather than Time-to-First-Token, and highlights the importance of turn detection and interruption handling.

### Amp — Agents in Orbs

- Company: Amp (ampcode.com)
- Announced: 2026-06-30
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/agents-in-orbs
- Record: https://forck.live/items/7734-agents-in-orbs
- Subject: Amp

Amp announced that its agents can now be launched remotely in 'orbs' — machines where agents run without supervision. Orbs are billed by the minute, start quickly, and go to sleep when no longer needed. This allows users to run multiple agents simultaneously without local resource conflicts.

### Anthropic — Introducing Claude Sonnet 5

- Company: Anthropic (anthropic.com)
- Announced: 2026-06-30
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.anthropic.com/news/claude-sonnet-5
- Record: https://forck.live/items/3252-introducing-claude-sonnet-5
- Subject: Claude

Anthropic introduced Claude Sonnet 5, a new model with frontier performance in coding, agents, and professional work at scale.

### Anthropic — Redeploying Fable 5

- Company: Anthropic (anthropic.com)
- Announced: 2026-06-30
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.anthropic.com/news/redeploying-fable-5
- Record: https://forck.live/items/3251-redeploying-fable-5
- Subject: Claude

Anthropic announces that Fable 5 returns globally on July 1, and also proposes an industry-wide framework for scoring jailbreak severity with partners including Amazon, Microsoft, Google, and other Glasswing partners.

### Cursor — MCPs and Organizations in Team Marketplaces

- Company: Cursor (cursor.com)
- Announced: 2026-06-30
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/team-marketplace-updates
- Record: https://forck.live/items/2593-mcps-and-organizations-in-team-marketplaces
- Subject: Cursor

Cursor's team marketplace now supports Team MCPs (admins can configure and distribute MCP servers) and organization groups for restricting marketplace access.

### Together AI — Together AI at ICML 2026: frontier research across the full stack

- Company: Together AI (together.ai)
- Announced: 2026-06-30
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/icml-2026
- Record: https://forck.live/items/2339-together-ai-at-icml-2026-frontier-research-across-the-full-stack
- Subject: Inference platform

Together AI announced nine research papers accepted at ICML 2026, covering topics across the full AI stack including frontier agents, model shaping, algorithmic optimizations, systems optimizations, and kernels. The papers include DSGym, ThunderAgent, TTT-Discover, RARO, V1, Aurora, Untied Ulysses, OEA, and ParallelKernelBench. The TTT-Discover paper uses an open 120B model (gpt-oss-120b) and achieves state-of-the-art discoveries in mathematics, GPU kernels, competitive algorithms, and biology.

### Hugging Face — Featuring Every Eval Ever Results on Hugging Face Model Pages

- Company: Hugging Face (huggingface.co)
- Announced: 2026-06-30
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/eee-community-evals
- Record: https://forck.live/items/1470-featuring-every-eval-ever-results-on-hugging-face-model-pages
- Subject: Platform

Every Eval Ever (EEE) and Hugging Face Community Evals are now intercompatible, enabling cross-posting and interpretation of evaluation results with a unified standardized metadata store. A converter allows EEE records to be sent to Community Evals, and results now carry a source badge linking to the full EEE record.

### OpenAI — Introducing GeneBench-Pro

- Company: OpenAI (openai.com)
- Announced: 2026-06-30
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-genebench-pro
- Record: https://forck.live/items/48-introducing-genebench-pro
- Subject: GPT / ChatGPT / API

Introduces GeneBench-Pro, a new benchmark for AI performance in genomics, biology, and scientific research using complex, real-world datasets.

### OpenAI — Core dump epidemiology: fixing an 18-year-old bug

- Company: OpenAI (openai.com)
- Announced: 2026-06-30
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/core-dump-epidemiology-data-infrastructure-bug
- Record: https://forck.live/items/46-core-dump-epidemiology-fixing-an-18-year-old-bug
- Subject: GPT / ChatGPT / API

OpenAI engineers used large-scale core dump analysis to debug rare infrastructure crashes, uncovering both a hardware fault and a long-standing software bug.

### Microsoft Research — Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity

- Company: Microsoft Research (microsoft.com)
- Announced: 2026-06-29T21:14:22+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.microsoft.com/en-us/research/blog/memora-a-harmonic-memory-representation-balancing-abstraction-and-specificity/
- Record: https://forck.live/items/2462-memora-a-harmonic-memory-representation-balancing-abstraction-and-specificity
- Subject: Research / Phi

Memora is a scalable memory system for long-horizon AI agents that decouples rich memory content from lightweight retrieval, achieving state-of-the-art results on LoCoMo and LongMemEval benchmarks while using up to 98% fewer context tokens. The paper is published at ICML 2026 and code is available on GitHub.

### Microsoft — Claude in Microsoft Foundry is now generally available

- Company: Microsoft (microsoft.com)
- Announced: 2026-06-29T17:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://azure.microsoft.com/en-us/blog/claude-in-microsoft-foundry-is-now-generally-available/
- Record: https://forck.live/items/2454-claude-in-microsoft-foundry-is-now-generally-available
- Subject: Azure AI

Claude in Microsoft Foundry is now generally available, hosted on Azure, providing enterprises with a production path that includes Azure-native controls, simplified procurement, and faster time to value.

### fal — The operational reality of marketing inside an AI company

- Company: fal (fal.ai)
- Announced: 2026-06-29T15:54:29+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.fal.ai/the-operational-reality-of-marketing-inside-an-ai-company/
- Record: https://forck.live/items/7996-the-operational-reality-of-marketing-inside-an-ai-company
- Subject: fal platform

The marketing and growth lead at fal describes how the company uses AI selectively in marketing workflows for tasks such as production mockups, creative exploration, and quickly visualizing ideas, rather than relying on full automation. She argues that the most useful applications of AI in marketing are narrow, practical, and controllable, and that the real value lies in reducing the cost of moving from instinct to artifact.

### Vercel — Vercel Open Source Program: Spring 2026 cohort

- Company: Vercel (vercel.com)
- Announced: 2026-06-29T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/vercel-open-source-program-spring-2026-cohort
- Record: https://forck.live/items/20693-vercel-open-source-program-spring-2026-cohort
- Subject: v0 / AI SDK / AI Gateway

Vercel announced the Spring 2026 cohort of its Open Source Program, which provides 17 open-source projects with $3,600 in Vercel platform credits, a starter pack with third-party service credits, and guidance from Vercel and the community. The program runs four times a year and aims to support maintainers in shipping and growing their projects.

### Runway — Runway and MIXI Announce Strategic Partnership

- Company: Runway (runway.com)
- Announced: 2026-06-29T10:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/runway-and-mixi-announce-strategic-partnership
- Record: https://forck.live/items/7871-runway-and-mixi-announce-strategic-partnership
- Subject: Gen / Aleph

Runway announced an enterprise partnership with MIXI, a Japanese technology and entertainment company. MIXI has deployed Runway across its creative operations, achieving significant time savings in video production. The partnership follows Runway's expansion in Japan, including opening a Tokyo office and committing $40 million to build its presence in the country.

### Vercel — Build realtime voice agents on AI Gateway

- Company: Vercel (vercel.com)
- Announced: 2026-06-29T07:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/realtime-voice-agents-on-ai-gateway
- Record: https://forck.live/items/20694-build-realtime-voice-agents-on-ai-gateway
- Subject: v0 / AI SDK / AI Gateway

Vercel added audio and voice support to AI Gateway, including realtime voice, text-to-speech, and speech-to-text capabilities. The new features are in beta and available in AI SDK 7, with initial models from OpenAI and xAI. Audio calls receive the same provider routing, observability, spend controls, and bring-your-own-key support as other modalities.

### OpenAI — Mapping Europe’s AI Workforce Opportunity

- Company: OpenAI (openai.com)
- Announced: 2026-06-29T07:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/mapping-ai-jobs-transition-eu
- Record: https://forck.live/items/49-mapping-europe-s-ai-workforce-opportunity
- Subject: GPT / ChatGPT / API

OpenAI published a report that analyzes the potential impact of AI on jobs in the European Union, focusing on occupations that may see automation, growth, or workflow changes.

### Ai2 — DiScoFormer: One transformer for density and score, across distributions

- Company: Ai2 (allenai.org)
- Announced: 2026-06-29
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://allenai.org/blog/discoformer
- Record: https://forck.live/items/18548-discoformer-one-transformer-for-density-and-score-across-distributions
- Subject: Olmo / Molmo / Asta

Ai2 introduced DiScoFormer, a transformer that estimates both the density and score of a distribution from a set of data points in a single forward pass without retraining. It outperforms kernel density estimation in high dimensions, cutting score error by about 6.5x and density error by more than 37x in 100 dimensions, and generalizes to out-of-distribution inputs via a label-free consistency loss.

### Cognition — Devin Fusion: Frontier Performance at 60% Lower Cost

- Company: Cognition (cognition.com)
- Announced: 2026-06-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-fusion
- Record: https://forck.live/items/7652-devin-fusion-frontier-performance-at-60-lower-cost
- Subject: Devin / SWE

Cognition introduced Devin Fusion, a multi-model harness that uses a frontier model and a cost-effective 'sidekick' model in parallel, achieving up to 60% cost reduction on the FrontierCode benchmark while maintaining performance. The system is available in preview on Devin.

### Cursor — Cursor Mobile App for iOS

- Company: Cursor (cursor.com)
- Announced: 2026-06-29
- Category: availability-change
- Coverage: 1 outlet
- Announcement: yes
- Group: covered
- Source: https://cursor.com/changelog/ios-mobile-app
- Record: https://forck.live/items/2594-cursor-mobile-app-for-ios
- Subject: Cursor

Cursor has released a mobile app for iOS in public beta, allowing users to launch cloud agents, use remote control, and receive push notifications, available to all paid plans.