# forck.live — 30 March – 5 April 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-W14, 30 March – 5 April 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-W14
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Google Research — Evaluating alignment of behavioral dispositions in LLMs

- Company: Google Research (research.google)
- Announced: 2026-04-03T08:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/evaluating-alignment-of-behavioral-dispositions-in-llms/
- Record: https://forck.live/items/1350-evaluating-alignment-of-behavioral-dispositions-in-llms
- Subject: Research

Google Research introduces a systematic evaluation framework that converts psychological questionnaires into situational judgment tests to measure how closely LLMs' behavioral dispositions align with human consensus. Analysis of 25 LLMs reveals gaps where model behavior deviates from human agreement or fails to capture the range of human opinions.

### Gradium — Gradium Phonon: On-Device TTS for Consumer Apps, NPCs, and Offline Products

- Company: Gradium (gradium.ai)
- Announced: 2026-04-03
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://gradium.ai/blog/gradium-phonon
- Record: https://forck.live/items/18324-gradium-phonon-on-device-tts-for-consumer-apps-npcs-and-offline-products
- Subject: Gradium TTS / Phonon

Gradium announced Phonon, an on-device text-to-speech model that runs entirely on CPU across Android and iOS with no network connection required. Each model is finetuned for one voice, one target language, and one content type, and is shipped as a licensed binary inside the app. Phonon is available through a private beta.

### Genspark — How CBRE Wins Competitive Listings with Genspark

- Company: Genspark (genspark.ai)
- Announced: 2026-04-03
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/cbre-genspark-case-study
- Record: https://forck.live/items/18258-how-cbre-wins-competitive-listings-with-genspark
- Subject: Genspark

### Together AI — AI for Systems: Using LLMs to Optimize Database Query Execution

- Company: Together AI (together.ai)
- Announced: 2026-04-03
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/using-llms-to-optimize-database-query-execution
- Record: https://forck.live/items/2362-ai-for-systems-using-llms-to-optimize-database-query-execution
- Subject: Inference platform

Together AI, Stanford University, the University of Wisconsin–Madison, and Bauplan collaborated to test whether LLMs can optimize database query execution plans. They introduce DBPlanBench, a harness that exposes the internal physical operator graph of Apache DataFusion to an LLM, and show that LLM-guided plan rewrites can improve execution performance without modifying the database engine itself.

### Together AI — Wan 2.7 video model suite now available on Together AI

- Company: Together AI (together.ai)
- Announced: 2026-04-03
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.together.ai/blog/wan-2-7-now-available-on-together-ai
- Record: https://forck.live/items/2361-wan-2-7-video-model-suite-now-available-on-together-ai
- Subject: Inference platform

Together AI announces the availability of the Wan 2.7 video model suite, beginning with text-to-video and soon including image-to-video, reference-to-video, and video edit. The suite offers features like audio-driven generation, frame-level conditioning, and flexible resolutions up to 1080P. Pricing starts at $0.10 per second of generated video.

### Google DeepMind — Gemma 4: Byte for byte, the most capable open models

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-04-02T16:00:49+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://deepmind.google/blog/gemma-4-byte-for-byte-the-most-capable-open-models/
- Record: https://forck.live/items/1242-gemma-4-byte-for-byte-the-most-capable-open-models
- Subject: Gemini

Google DeepMind announces Gemma 4, its most intelligent open models to date, designed for advanced reasoning and agentic workflows.

### Sarvam AI — Evaluating Indian Language ASR

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-04-02T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.sarvam.ai/blogs/evaluating-indian-language-asr/
- Record: https://forck.live/items/18387-evaluating-indian-language-asr
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI published a guide on evaluating Indian language ASR systems, introducing LLM-based metrics (LLM-WER, LLM-CER, Intent Score, Entity Preservation Score) and open-source frameworks to address limitations of traditional metrics like WER and CER for Indian languages.

### OpenAI — OpenAI acquires TBPN

- Company: OpenAI (openai.com)
- Announced: 2026-04-02T10:30:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/openai-acquires-tbpn
- Record: https://forck.live/items/213-openai-acquires-tbpn
- Subject: GPT / ChatGPT / API

OpenAI announces the acquisition of TBPN to accelerate global conversations around AI and support independent media.

### OpenAI — Codex now offers more flexible pricing for teams

- Company: OpenAI (openai.com)
- Announced: 2026-04-02T10:00:00+00:00
- Category: pricing-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/codex-flexible-pricing-for-teams
- Record: https://forck.live/items/214-codex-now-offers-more-flexible-pricing-for-teams
- Subject: GPT / ChatGPT / API

OpenAI announces pay-as-you-go pricing for Codex for ChatGPT Business and Enterprise teams.

### Vercel — Optimizing Vercel Sandbox snapshots

- Company: Vercel (vercel.com)
- Announced: 2026-04-02T04:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/optimizing-vercel-sandbox-snapshots
- Record: https://forck.live/items/20720-optimizing-vercel-sandbox-snapshots
- Subject: v0 / AI SDK / AI Gateway

Vercel optimized filesystem snapshot restores in Vercel Sandbox, reducing p75 restore time from over 40 seconds to under one second through parallelization and local caching. The optimizations also power Automatic Persistence, now in beta, which automatically snapshots and restores filesystems on stop and resume.

### Supabase — 100,000 GitHub stars

- Company: Supabase (supabase.com)
- Announced: 2026-04-02
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/100000-github-stars
- Record: https://forck.live/items/21295-100-000-github-stars
- Subject: Supabase

Supabase announced that its GitHub repository has received 100,000 stars and that eight million developers now build with the platform. The blog post reflects on the community that contributed to this milestone.

### Upstage — Upstage AI Launches AI-Powered Agentic Workflow Solution for Document-Heavy Processes

- Company: Upstage (upstage.ai)
- Announced: 2026-04-02
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/blog/en/upstage-ai-launches-ai-powered-agentic-workflow-solution-for-document-heavy-processes
- Record: https://forck.live/items/18455-upstage-ai-launches-ai-powered-agentic-workflow-solution-for-document-heavy
- Subject: Solar Pro / Solar Mini

Announcements Upstage Team Announcements Read Blog

### Krea — Seedance 2.0: now on Krea

- Company: Krea (krea.ai)
- Announced: 2026-04-02
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.krea.ai/blog/seedance-2
- Record: https://forck.live/items/15083-seedance-2-0-now-on-krea
- Subject: Krea 2 / AI creative platform

Krea has made Seedance 2.0 available on all paid plans, describing it as the strongest AI video model it has tested for motion coherence, camera control, and clip-level stability. The model maintains full-clip coherence, directional camera pans, and character recognition across frames, producing output stable enough for editing and compositing. Users can access it through the Krea video tool's model selector.

### Microsoft AI — Today we're announcing 3 new world class MAI models, available in Foundry

- Company: Microsoft AI (microsoft.ai)
- Announced: 2026-04-02
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://microsoft.ai/news/today-were-announcing-3-new-world-class-mai-models-available-in-foundry/
- Record: https://forck.live/items/10927-today-we-re-announcing-3-new-world-class-mai-models-available-in-foundry
- Subject: MAI

Microsoft AI announced three new MAI models available in Microsoft Foundry and MAI Playground. MAI-Transcribe-1 offers speech-to-text transcription across 25 languages with batch speeds 2.5x faster than Azure Fast. MAI-Voice-1 generates natural speech with emotional range and can create custom voices from seconds of audio. MAI-Image-2 provides at least 2x faster image generation on Foundry and Copilot compared to previous versions.

### Microsoft AI — State of the Art Speech Recognition with MAI-Transcribe-1

- Company: Microsoft AI (microsoft.ai)
- Announced: 2026-04-02
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://microsoft.ai/news/state-of-the-art-speech-recognition-with-mai-transcribe-1
- Record: https://forck.live/items/8431-state-of-the-art-speech-recognition-with-mai-transcribe-1
- Subject: MAI

Microsoft AI announced MAI-Transcribe-1, a multilingual speech-to-text model that achieves the lowest Word Error Rate on FLEURS across 25 languages, delivers batch transcription 2.5x faster than the current Azure Fast offering, and is priced at $0.36 per hour of audio. The model is now in public preview on Microsoft Foundry and is being rolled out to Copilot Voice mode and Microsoft Teams.

### Microsoft AI — Announcing 3 new world class MAI models, available in Foundry

- Company: Microsoft AI (microsoft.ai)
- Announced: 2026-04-02
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://microsoft.ai/news/today-were-announcing-3-new-world-class-mai-models-available-in-foundry
- Record: https://forck.live/items/8430-announcing-3-new-world-class-mai-models-available-in-foundry
- Subject: MAI

Microsoft AI announced three new MAI models: MAI-Transcribe-1 for speech-to-text, MAI-Voice-1 for voice generation, and MAI-Image-2 for image generation. The models are available in Microsoft Foundry and MAI Playground with competitive pricing. MAI-Transcribe-1 offers state-of-the-art transcription across the top 25 most-used languages, MAI-Voice-1 can generate 60 seconds of audio in one second, and MAI-Image-2 provides at least 2x faster generation times.

### Cursor — New Cursor Interface

- Company: Cursor (cursor.com)
- Announced: 2026-04-02
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/3-0
- Record: https://forck.live/items/2623-new-cursor-interface
- Subject: Cursor

Cursor 3 introduces a new interface with an Agents Window for running multiple agents in parallel across different environments, a Design Mode for annotating UI elements in the browser, Agent Tabs for viewing multiple chats side-by-side, and various other improvements and bug fixes.

### Together AI — Deepgram speech-to-text and voice models now available natively on Together AI

- Company: Together AI (together.ai)
- Announced: 2026-04-02
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/deepgram-speech-to-text-and-voice-models-now-available-natively-on-together-ai
- Record: https://forck.live/items/2363-deepgram-speech-to-text-and-voice-models-now-available-natively-on-together-ai
- Subject: Inference platform

Deepgram's speech-to-text and voice models (Nova-3, Nova-3 Multilingual, Flux, Aura-2) are now available natively on Together AI's Dedicated Model Inference platform, enabling teams to run the full voice pipeline (STT, LLM, TTS) on one production platform with enterprise controls.

### Hugging Face — Welcome Gemma 4: Frontier multimodal intelligence on device

- Company: Hugging Face (huggingface.co)
- Announced: 2026-04-02
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/gemma4
- Record: https://forck.live/items/1524-welcome-gemma-4-frontier-multimodal-intelligence-on-device
- Subject: Platform

Google DeepMind has released the Gemma 4 family of multimodal models on Hugging Face. The models are open-weight under the Apache 2 license and support text, image, and audio inputs with long context windows ranging from 128K to 256K tokens. The family includes five sizes: E2B, E4B, 12B Unified, 31B, and 26B A4B, each available in base and instruction-tuned versions.

### Lovable — A Bug Hunt in Our Kubernetes Cluster

- Company: Lovable (lovable.dev)
- Announced: 2026-04-01T22:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/hunting-networking-bugs-in-kubernetes
- Record: https://forck.live/items/12092-a-bug-hunt-in-our-kubernetes-cluster
- Subject: Lovable / AI app builder

Lovable engineers debugged two layered networking failures in their Google Kubernetes Engine cluster. First, a concurrent map-access panic in Google's anetd WireGuard integration code caused pod crashes. After disabling transparent node-to-node encryption, a second issue emerged: an MTU mismatch between nodes that had and had not restarted, causing connection failures to Valkey. The fix required rolling all nodes to ensure consistent MTU configuration.

### Alibaba — Qwen3.6-Plus: Towards Real World Agents

- Company: Alibaba (alibaba.com)
- Announced: 2026-04-01T20:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwen3.6
- Record: https://forck.live/items/4602-qwen3-6-plus-towards-real-world-agents
- Subject: Qwen

Alibaba announces Qwen3.6-Plus, a new model with enhanced agentic coding capabilities, multimodal reasoning, and a 1M context window, available immediately via API.

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

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-04-01T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-march-2026/
- Record: https://forck.live/items/1317-the-latest-ai-news-we-announced-in-march-2026
- Subject: Gemini

Google posted an overview of their AI updates from March 2026, but no specific details about models, products, or other announcements were provided in the source text.

### TII — Falcon Perception

- Company: TII (tii.ae)
- Announced: 2026-04-01T07:13:20+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/tiiuae/falcon-perception
- Record: https://forck.live/items/1525-falcon-perception

TII introduces Falcon Perception, a 0.6B-parameter early-fusion Transformer for open-vocabulary grounding and segmentation, and Falcon OCR, a 0.3B-parameter OCR model. Falcon Perception achieves 68.0 Macro-F1 on SA-Co, beating SAM 3's 62.3. Falcon OCR achieves 80.3 on olmOCR and 88.6 on OmniDocBench with high throughput. They also introduce PBench, a diagnostic benchmark.

### Vercel — How Waldium made a blog platform work for humans and AI alike

- Company: Vercel (vercel.com)
- Announced: 2026-04-01T05:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-waldium-made-a-blog-platform-work-for-humans-and-ai-alike
- Record: https://forck.live/items/20721-how-waldium-made-a-blog-platform-work-for-humans-and-ai-alike
- Subject: v0 / AI SDK / AI Gateway

Vercel's case study documents how Waldium, a two-person YC-backed startup, built an agentic CMS on Vercel that gives each customer blog its own MCP server endpoint. The platform serves 500+ customer blogs from a single Next.js deployment, with AI query response times under 50ms globally and 45% lower infrastructure cost versus a per-customer deployment model. One customer, Saphira AI, produced over 1,000 posts in a month, cutting research time from weeks to hours.

### OpenAI — Gradient Labs gives every bank customer an AI account manager

- Company: OpenAI (openai.com)
- Announced: 2026-04-01T02:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/gradient-labs
- Record: https://forck.live/items/215-gradient-labs-gives-every-bank-customer-an-ai-account-manager
- Subject: GPT / ChatGPT / API

Gradient Labs uses GPT-4.1 and GPT-5.4 mini and nano to power AI agents that automate banking support workflows with low latency and high reliability.

### Ant Ling — Taking the Pulse of Agentic AI from the Developer Community at the End of Q1 2026

- Company: Ant Ling (ant-ling.com)
- Announced: 2026-04-01
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://developer.ant-ling.com/en/blogs/agentic-landscape-in-2026-Q1
- Record: https://forck.live/items/22332-taking-the-pulse-of-agentic-ai-from-the-developer-community-at-the-end-of-q1
- Subject: Ling / Ring / Ming

Today, I want to share some observations on the Agentic AI ecosystem from the vantage point of 2026’s first quarter—technical trends read from popular projects, portraits of AI developers, and the subtle relationship between developers and AI tools.

### Supabase — Supabase docs over SSH

- Company: Supabase (supabase.com)
- Announced: 2026-04-01
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/supabase-docs-over-ssh
- Record: https://forck.live/items/21296-supabase-docs-over-ssh
- Subject: Supabase

Supabase launched supabase.sh, a public SSH server that provides AI coding agents with bash access to Supabase documentation as markdown files in a virtual file system. The service is free and available now as a public experiment, allowing agents to browse, search, and read docs using standard bash commands.

### Factory AI — Legacy-Bench: Can AI Agents Maintain the World's Most Critical Software?

- Company: Factory AI (factory.com)
- Announced: 2026-04-01
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/legacy-bench
- Record: https://forck.live/items/17855-legacy-bench-can-ai-agents-maintain-the-world-s-most-critical-software
- Subject: Droid

Factory AI released Legacy-Bench, a benchmark of hundreds of tasks in COBOL, Fortran, Java 7, and other legacy languages, designed to measure frontier AI agent capabilities on legacy software engineering. Across 12 model-agent combinations, overall pass rates ranged from 16.9% to 42.5%, compared to >70% on modern benchmarks. Performance varied sharply by language and task type, with COBOL proving the hardest and no single model winning across all categories.

### Luma — Introducing Team Members and Admins on Your Luma Account

- Company: Luma (lumalabs.ai)
- Announced: 2026-04-01
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lumalabs.ai/news/luma-teams
- Record: https://forck.live/items/7969-introducing-team-members-and-admins-on-your-luma-account
- Subject: Ray / Dream Machine

Luma introduced the ability to add team members and admins to Luma accounts, giving each member their own workspace and login, with two roles (Admin and Team Member) and a usage dashboard for admins. This feature is available across all Luma plans.

### Z.ai — GLM-5V-Turbo

- Company: Z.ai (z.ai)
- Announced: 2026-04-01
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://docs.z.ai/release-notes/new-released#2026-04-01
- Record: https://forck.live/items/4449-glm-5v-turbo
- Subject: GLM

Z.ai released GLM-5V-Turbo, a new multimodal model designed for vision-based coding, GUI task execution, and Agent workflows. It supports native understanding of images, video, and text, and enhances tool use and end-to-end action planning.

### Together AI — Inside the Together AI kernels team

- Company: Together AI (together.ai)
- Announced: 2026-04-01
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/inside-the-together-ai-kernels-team
- Record: https://forck.live/items/2364-inside-the-together-ai-kernels-team
- Subject: Inference platform

The blog post tells the story of Together AI's kernels team, highlighting their work on FlashAttention, the ThunderKittens library, and their rapid optimization of kernels for NVIDIA's Blackwell GPUs, achieving up to 2x speedups over cuBLAS on H100s within a week of hardware access.

### Hugging Face — Any Custom Frontend with Gradio's Backend

- Company: Hugging Face (huggingface.co)
- Announced: 2026-04-01
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/introducing-gradio-server
- Record: https://forck.live/items/1526-any-custom-frontend-with-gradio-s-backend
- Subject: Platform

Hugging Face announces gradio.Server, a new server class that extends FastAPI to allow developers to build custom frontends with any framework (React, Svelte, vanilla HTML/JS) while leveraging Gradio's queuing, concurrency, ZeroGPU, and API infrastructure.

### Google Research — Building better AI benchmarks: How many raters are enough?

- Company: Google Research (research.google)
- Announced: 2026-03-31T16:16:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://research.google/blog/building-better-ai-benchmarks-how-many-raters-are-enough/
- Record: https://forck.live/items/1351-building-better-ai-benchmarks-how-many-raters-are-enough
- Subject: Research

Google Research introduces an evaluation framework for ML models that optimizes the trade-off between the number of items and raters per item to build reproducible AI benchmarks that capture human disagreement. The research finds that the common practice of 3-5 raters is often insufficient and that more than 10 raters per item is often needed. The simulator has been open-sourced on GitHub.

### ByteDance — ByteDance Seed 2027 Foundation Model Campus Recruitment is Now Open (Internships Included)

- Company: ByteDance (bytedance.com)
- Announced: 2026-03-31T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://seed.bytedance.com/en/blog/bytedance-seed-2027-foundation-model-campus-recruitment-is-now-open-internships-included
- Record: https://forck.live/items/4666-bytedance-seed-2027-foundation-model-campus-recruitment-is-now-open
- Subject: Seed / Doubao

ByteDance Seed announces campus recruitment for 2027 graduates and interns, focusing on foundation models, visual intelligence, speech intelligence, machine learning systems, LLM applications, and embodied intelligence.

### Hugging Face — Granite 4.0 3B Vision: Compact Multimodal Intelligence for Enterprise Documents

- Company: Hugging Face (huggingface.co)
- Announced: 2026-03-31T15:10:41+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/ibm-granite/granite-4-vision
- Record: https://forck.live/items/1527-granite-4-0-3b-vision-compact-multimodal-intelligence-for-enterprise-documents
- Subject: Platform

IBM and Hugging Face announce Granite 4.0 3B Vision, a compact vision-language model for enterprise document understanding, featuring table extraction, chart understanding, and semantic KVP extraction. It is built as a LoRA adapter on Granite 4.0 Micro, using the ChartNet dataset and DeepStack architecture.

### OpenAI — Accelerating the next phase of AI

- Company: OpenAI (openai.com)
- Announced: 2026-03-31T13:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/accelerating-the-next-phase-ai
- Record: https://forck.live/items/216-accelerating-the-next-phase-of-ai
- Subject: GPT / ChatGPT / API

OpenAI raises $122 billion in new funding to expand frontier AI globally, invest in next-generation compute, and meet growing demand for ChatGPT, Codex, and enterprise AI.

### Mistral AI — Spaces: A CLI Built for Humans and Agents

- Company: Mistral AI (mistral.ai)
- Announced: 2026-03-31T12:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/spaces/
- Record: https://forck.live/items/9934-spaces-a-cli-built-for-humans-and-agents
- Subject: Mistral / Le Chat

Mistral AI released Spaces, a CLI tool that scaffolds projects, spins up dev environments, generates configs, and deploys to staging. The tool is designed for both human developers and coding agents, with every interactive prompt having a flag equivalent so agents can operate autonomously. Spaces also generates context.json and AGENTS.md files to help agents understand the project structure.

### Runway — Introducing Runway Builders

- Company: Runway (runway.com)
- Announced: 2026-03-31T12:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/introducing-runway-builders
- Record: https://forck.live/items/7890-introducing-runway-builders
- Subject: Gen / Aleph

Runway launched Runway Builders, a program providing Seed to Series C startups with free API credits and early access to Runway Characters, a real-time video agent API powered by GWM-1.

### ElevenLabs — Scaling multilingual diplomacy during the Polish presidency of the Council of the EU

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-03-31T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://elevenlabs.io/blog/scaling-multilingual-diplomacy-during-the-polish-presidency-of-the-council-of-the-eu
- Record: https://forck.live/items/3368-scaling-multilingual-diplomacy-during-the-polish-presidency-of-the-council-of
- Subject: Voice AI

### Hugging Face — Training mRNA Language Models Across 25 Species for $165

- Company: Hugging Face (huggingface.co)
- Announced: 2026-03-31T08:23:44+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/OpenMed/training-mrna-models-25-species
- Record: https://forck.live/items/1528-training-mrna-language-models-across-25-species-for-165
- Subject: Platform

OpenMed built an end-to-end protein AI pipeline covering structure prediction, sequence design, and codon optimization. They trained and compared multiple transformer architectures for codon-level language modeling, finding CodonRoBERTa-large-v2 as the best with perplexity 4.10 and Spearman CAI correlation 0.40. They scaled to 25 species, training 4 production models in 55 GPU-hours, and built a species-conditioned system that no other open-source project offers.

### Vercel — How FLORA shipped a creative agent on Vercel's AI stack

- Company: Vercel (vercel.com)
- Announced: 2026-03-31T04:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-flora-shipped-a-creative-agent-on-vercels-ai-stack
- Record: https://forck.live/items/20722-how-flora-shipped-a-creative-agent-on-vercel-s-ai-stack
- Subject: v0 / AI SDK / AI Gateway

FLORA built a creative agent called FAUNA that generates visual explorations from user ideas. The agent uses Vercel's AI SDK, Workflow SDK, and Fluid compute to orchestrate parallel image generation across 50+ models. FLORA reported 2x faster time to production and eliminated infrastructure debates after migrating from LangChain and Temporal.

### Google Research — Safeguarding cryptocurrency by disclosing quantum vulnerabilities responsibly

- Company: Google Research (research.google)
- Announced: 2026-03-31T02:03:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/safeguarding-cryptocurrency-by-disclosing-quantum-vulnerabilities-responsibly/
- Record: https://forck.live/items/1352-safeguarding-cryptocurrency-by-disclosing-quantum-vulnerabilities-responsibly
- Subject: Research

Google Research publishes a whitepaper with updated quantum resource estimates for breaking elliptic curve cryptography, showing a 20-fold reduction in physical qubits needed, and proposes a responsible disclosure method using zero-knowledge proofs, urging the cryptocurrency community to transition to post-quantum cryptography.

### Liquid AI — LFM2.5-350M: No Size Left Behind

- Company: Liquid AI (liquid.ai)
- Announced: 2026-03-31
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.liquid.ai/blog/lfm2-5-350m-no-size-left-behind
- Record: https://forck.live/items/16785-lfm2-5-350m-no-size-left-behind
- Subject: LFM / d1 models

Liquid AI released LFM2.5-350M, an improved version of its 350M model with additional pre-training (from 10T to 28T tokens) and large-scale reinforcement learning. The model outperforms models more than twice its size on benchmarks for knowledge, instruction following, data extraction, and tool use, and achieves a peak throughput of 40.4K output tokens per second on a single H100 GPU. It is available on Hugging Face, LEAP, and the Playground, with day-one support for multiple inference engines and expanded hardware and software partnerships.

### Together AI — Aurora

- Company: Together AI (together.ai)
- Announced: 2026-03-31
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/aurora
- Record: https://forck.live/items/2365-aurora
- Subject: Inference platform

Aurora is an open-source, RL-based framework for adaptive speculative decoding that learns from live inference traces and continuously updates the speculator without interrupting serving, achieving an additional 1.25x speedup over a well-trained static speculator.

### Hugging Face — TRL v1.0: Post-Training Library Built to Move with the Field

- Company: Hugging Face (huggingface.co)
- Announced: 2026-03-31
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/trl-v1
- Record: https://forck.live/items/1529-trl-v1-0-post-training-library-built-to-move-with-the-field
- Subject: Platform

Hugging Face releases TRL v1.0, marking a shift from a research codebase to a stable library for post-training methods, now implementing over 75 methods with a stable/experimental split and semantic versioning.

### Vercel — Agent responsibly

- Company: Vercel (vercel.com)
- Announced: 2026-03-30T05:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/agent-responsibly
- Record: https://forck.live/items/20723-agent-responsibly
- Subject: v0 / AI SDK / AI Gateway

Vercel published a framework for safely deploying agent-generated code, advocating for self-driving deployments, continuous validation, and executable guardrails rather than relying solely on code review.

### Vercel — Making Turborepo 96% faster with agents, sandboxes, and humans

- Company: Vercel (vercel.com)
- Announced: 2026-03-30T04:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/making-turborepo-ninety-six-percent-faster-with-agents-sandboxes-and-humans
- Record: https://forck.live/items/20724-making-turborepo-96-faster-with-agents-sandboxes-and-humans
- Subject: v0 / AI SDK / AI Gateway

Vercel announced that Turborepo's task graph computation is now 81-91% faster, with Time to First Task 11x faster on its 1,000+ package monorepo. The performance gains, achieved through a combination of AI agents, Vercel Sandboxes, and engineering practices, can reach up to 96% depending on repository size and complexity.

### Inception — Introducing Mercury Edit 2

- Company: Inception (inceptionlabs.ai)
- Announced: 2026-03-30
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.inceptionlabs.ai/blog/introducing-mercury-edit-2
- Record: https://forck.live/items/18522-introducing-mercury-edit-2
- Subject: Mercury

Inception released Mercury Edit 2, a diffusion LLM for next-edit prediction in code editors. The model's edits are accepted 48% more often and it is 27% more selective in displayed edits compared to its predecessor. It is available on the Inception Platform at $0.25 per million input tokens and $0.75 per million output tokens.

### Amp — Amp Free Is Ad-Free

- Company: Amp (ampcode.com)
- Announced: 2026-03-30
- Category: pricing-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/amp-free-is-ad-free
- Record: https://forck.live/items/7752-amp-free-is-ad-free
- Subject: Amp

Amp removed ads from its Amp Free tier, eliminating the ad-supported free tier. The company will also pause or reduce the $10 daily free usage grant for less-active users, making it more available for users on up-to-date versions and workflows.