# forck.live — 16–22 March 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-W12, 16–22 March 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-W12
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Replit — Live from Replit HQ: Agent 4 Launch Pt. 1

- Company: Replit (replit.com)
- Announced: 2026-03-20T21:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replit.com/blog/live-from-hq-agent4-launch-pt1
- Record: https://forck.live/items/17914-live-from-replit-hq-agent-4-launch-pt-1
- Subject: Replit Agent

Replit hosted a live stream from its headquarters to celebrate the launch of Agent 4, featuring demos of a taste-development app and a live 3D city built with the platform. Team members discussed how Agent 4 enables collaborative app building with Infinite Canvas, parallel tasks, and unified builds. The creator program was announced as open for applications.

### NVIDIA — Build a Domain-Specific Embedding Model in Under a Day

- Company: NVIDIA (nvidia.com)
- Announced: 2026-03-20T19:38:16+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/nvidia/domain-specific-embedding-finetune
- Record: https://forck.live/items/1532-build-a-domain-specific-embedding-model-in-under-a-day

NVIDIA and Hugging Face published a tutorial on fine-tuning a domain-specific embedding model in under a day using a single GPU, releasing a synthetic training dataset generated from NVIDIA's public documentation and open-source recipe integrations.

### Harvey — Your Data, Your Control: How Harvey Manages Customer Data

- Company: Harvey (harvey.ai)
- Announced: 2026-03-20T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/how-harvey-manages-customer-data
- Record: https://forck.live/items/17593-your-data-your-control-how-harvey-manages-customer-data
- Subject: Harvey

Sharing the core pillars that underpin our commitment to keeping customer data secure. Legal teams handle some of the most sensitive data in the world, and Harvey’s architecture was built with that responsibility at its core.

### Genspark — Seeing AGI (7): The Token Divide — Why Unlimited AI Access Is Now a Corporate Survival Imperative

- Company: Genspark (genspark.ai)
- Announced: 2026-03-20
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/seeing-agi-impact7
- Record: https://forck.live/items/18259-seeing-agi-7-the-token-divide-why-unlimited-ai-access-is-now-a-corporate
- Subject: Genspark

### LG AI Research — LG AI Insight 2026 A Field Report : Charting the Present and Future of AI

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-03-20
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/news/view?seq=635
- Record: https://forck.live/items/10993-lg-ai-insight-2026-a-field-report-charting-the-present-and-future-of-ai
- Subject: EXAONE

LG AI Research held LG AI Insight 2026 on March 19, an internal event to share recent research and real-world AI applications, with around 1,300 employees registered. The seminar featured presentations on EXAONE, Agentic AI, and a case study from the National Pension Service using EXAONE Data Foundry. A poster session covered 46 projects, and hands-on booths allowed attendees to experience ChatEXAONE.

### Cognition — Devin can now Schedule Devins

- Company: Cognition (cognition.com)
- Announced: 2026-03-20
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-can-now-schedule-devins
- Record: https://forck.live/items/7674-devin-can-now-schedule-devins
- Subject: Devin / SWE

Cognition announced that Devin can now schedule its own sessions to perform recurring tasks automatically. Users describe what should happen on a recurring basis, and Devin sets up the schedule and runs it. Devin carries state between runs, building on context from previous sessions, and can also compose with Managed Devins for parallel execution.

### fal — Connect your AI to 1,000+ models with the fal MCP Server

- Company: fal (fal.ai)
- Announced: 2026-03-19T20:14:52+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.fal.ai/connect-your-ai-to-1-000-models-with-the-fal-mcp-server/
- Record: https://forck.live/items/8001-connect-your-ai-to-1-000-models-with-the-fal-mcp-server
- Subject: fal platform

fal launched the fal MCP Server, a hosted endpoint that lets AI assistants search, run, and chain over 1,000 generative AI models directly from a conversation. The server exposes nine tools for model discovery, execution, and utility, and is free to use with standard fal pricing for model runs.

### Lovable — Go beyond building apps with Lovable

- Company: Lovable (lovable.dev)
- Announced: 2026-03-19T16:50:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/go-beyond-building-full-stack-apps-with-lovable
- Record: https://forck.live/items/12095-go-beyond-building-apps-with-lovable
- Subject: Lovable / AI app builder

Lovable's AI agent can now run code and execute scripts to handle data analysis, file processing, and document generation directly, enabling users to analyze files, create reports and presentations, turn files into working apps, and generate images and videos from the same conversation where they build products.

### Vercel — Two startups at global scale without DevOps

- Company: Vercel (vercel.com)
- Announced: 2026-03-19T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/two-startups-at-global-scale-without-devops
- Record: https://forck.live/items/20729-two-startups-at-global-scale-without-devops
- Subject: v0 / AI SDK / AI Gateway

Vercel's case study documents how Leonardo.AI and Relevance AI operate at global scale without dedicated DevOps teams, using Vercel's platform to handle infrastructure. Leonardo.AI processes over 4.5 million images daily and reduced build times from over 10 minutes to 2 minutes after switching to Vercel. Relevance AI runs 50,000 agents autonomously across systems like Salesforce and HubSpot with no infrastructure team.

### Vercel — Build knowledge agents without embeddings

- Company: Vercel (vercel.com)
- Announced: 2026-03-19T13:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/build-knowledge-agents-without-embeddings
- Record: https://forck.live/items/20728-build-knowledge-agents-without-embeddings
- Subject: v0 / AI SDK / AI Gateway

Vercel open-sourced the Knowledge Agent Template, a file-system-based agent that replaces vector databases and embeddings with bash commands in isolated sandboxes. The template uses Vercel Sandbox, AI SDK, and Chat SDK to deploy agents that search files with grep, find, and cat, providing deterministic and explainable results. Vercel reports that their sales call summarization agent went from ~$1.00 to ~$0.25 per call with improved output quality.

### Lovable — Prototype-to-production handoff: how non-technical teams use Lovable without bypassing engineering

- Company: Lovable (lovable.dev)
- Announced: 2026-03-19T10:02:09+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/prototype-to-production-handoff-how-non-technical-teams-use-lovable-without-bypassing-engineering
- Record: https://forck.live/items/12096-prototype-to-production-handoff-how-non-technical-teams-use-lovable-without
- Subject: Lovable / AI app builder

Lovable's guide describes a three-phase handoff process where non-technical teams build and validate prototypes in Lovable, export them to GitHub, and then engineers take over in their preferred tools. The process is designed so that prototypes do not ship directly to production, allowing engineering to retain control over what goes live. Lovable provides governance features including a security center, admin controls, and role-based access to manage non-technical builds.

### OpenAI — How we monitor internal coding agents for misalignment

- Company: OpenAI (openai.com)
- Announced: 2026-03-19T10:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/how-we-monitor-internal-coding-agents-misalignment
- Record: https://forck.live/items/225-how-we-monitor-internal-coding-agents-for-misalignment
- Subject: GPT / ChatGPT / API

OpenAI describes how it uses chain-of-thought monitoring to study misalignment in internal coding agents, analyzing real-world deployments to detect risks and strengthen AI safety safeguards.

### Vercel — Chat SDK brings agents to your users

- Company: Vercel (vercel.com)
- Announced: 2026-03-19T04:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/chat-sdk-brings-agents-to-your-users
- Record: https://forck.live/items/20730-chat-sdk-brings-agents-to-your-users
- Subject: v0 / AI SDK / AI Gateway

Vercel released Chat SDK, a TypeScript library that lets developers build bots for multiple chat platforms (Slack, Microsoft Teams, Google Chat, Discord, Telegram, GitHub, and Linear) from a single codebase. The SDK abstracts platform-specific API quirks into adapters, handles streaming, formatting, and element rendering (tables, cards, modals) per platform, and integrates with AI SDK for streaming LLM responses. It also automatically converts channel/user names and markdown to each platform's native format.

### Supabase — Supabase Joins the Stripe Projects Developer Preview

- Company: Supabase (supabase.com)
- Announced: 2026-03-19
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/supabase-joins-the-stripe-projects-developer-preview
- Record: https://forck.live/items/21298-supabase-joins-the-stripe-projects-developer-preview
- Subject: Supabase

Supabase has joined the Stripe Projects developer preview, allowing developers and AI agents to provision a complete Supabase project—including Postgres database, Auth, Storage, Edge Functions, and Realtime—with a single Stripe CLI command. The integration provides deterministic, repeatable provisioning and delivers credentials to the local environment or stores them in the Stripe Project for agent access. The database lives in the user's own Supabase account, and credentials can be rotated at any time.

### Thinking Machines Lab — Training LLMs to Predict World Events (Guest Post with Mantic)

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2026-03-19
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://thinkingmachines.ai/news/training-llms-to-predict-world-events
- Record: https://forck.live/items/18559-training-llms-to-predict-world-events-guest-post-with-mantic
- Subject: Thinking Machines / Inkling

Thinking Machines Lab and Mantic fine-tuned gpt-oss-120b on ~10,000 binary forecasting questions using reinforcement learning. The fine-tuned model achieved marginally superior performance to frontier LLMs in a head-to-head contest, despite much lower initial performance. In an optimal ensemble, the fine-tuned model and Grok 4 were the most important contributors.

### Upstage — AMD and Upstage Expand Strategic Collaboration to Advance Sovereign AI Infrastructure in Korea

- Company: Upstage (upstage.ai)
- Announced: 2026-03-19
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/news/upstage-amd-partnership
- Record: https://forck.live/items/18424-amd-and-upstage-expand-strategic-collaboration-to-advance-sovereign-ai
- Subject: Solar Pro / Solar Mini

AMD and Upstage expanded their strategic collaboration to accelerate AI model development and deployment in Korea. Upstage will deploy AMD Instinct MI355 GPUs over the next year to scale its large language models and document processing engines, and contribute to a government-led sovereign AI model project.

### Microsoft AI — Introducing MAI-Image-2

- Company: Microsoft AI (microsoft.ai)
- Announced: 2026-03-19
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://microsoft.ai/news/introducing-mai-image-2/
- Record: https://forck.live/items/10954-introducing-mai-image-2
- Subject: MAI

Microsoft AI announced MAI-Image-2, a text-to-image model ranked #3 on the Arena.ai leaderboard. The model features enhanced photorealism, reliable in-image text generation, and rich scene generation. It is available for preview in the MAI Playground, rolling out on Copilot and Bing Image Creator, with API access for select customers and planned availability on Microsoft Foundry.

### Microsoft AI — Introducing MAI-Image-2: for limitless creativity

- Company: Microsoft AI (microsoft.ai)
- Announced: 2026-03-19
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://microsoft.ai/news/introducing-mai-image-2
- Record: https://forck.live/items/8432-introducing-mai-image-2-for-limitless-creativity
- Subject: MAI

Microsoft AI announced MAI-Image-2, a text-to-image model that ranks third on the Arena.ai leaderboard. The model is available for preview in the MAI Playground and is rolling out on Copilot and Bing Image Creator, with API access for select customers and upcoming availability on Microsoft Foundry.

### Cognition — Devin can now Manage Devins

- Company: Cognition (cognition.com)
- Announced: 2026-03-19
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-can-now-manage-devins
- Record: https://forck.live/items/7675-devin-can-now-manage-devins
- Subject: Devin / SWE

Cognition announced that Devin, its AI software engineer, can now break down large tasks and delegate them to a team of managed Devins that work in parallel. Each managed Devin runs in its own isolated virtual machine with its own terminal, browser, and development environment. The main Devin session acts as a coordinator, scoping work, assigning pieces, monitoring progress, resolving conflicts, and compiling results.

### Cursor — Composer 2

- Company: Cursor (cursor.com)
- Announced: 2026-03-19
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/composer-2
- Record: https://forck.live/items/2625-composer-2
- Subject: Cursor

Cursor announces Composer 2, a new version of its coding assistant with improved performance on challenging coding tasks, and introduces two pricing tiers.

### OpenAI — OpenAI to acquire Astral

- Company: OpenAI (openai.com)
- Announced: 2026-03-19
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/openai-to-acquire-astral
- Record: https://forck.live/items/226-openai-to-acquire-astral
- Subject: GPT / ChatGPT / API

The text states that Codex growth is accelerated to power the next generation of Python developer tools.

### Alibaba — Qwen3.5-Max-Preview Now Available on Arena

- Company: Alibaba (alibaba.com)
- Announced: 2026-03-18T20:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://qwen.ai/blog?id=qwen3.5-max-preview
- Record: https://forck.live/items/4590-qwen3-5-max-preview-now-available-on-arena
- Subject: Qwen

Qwen3.5-Max-Preview has been deployed on LMSys Arena for preliminary evaluation, with a full release expected within two weeks.

### Harvey — Harvey’s Box Integration: Your Legal Documents, Now in Harvey

- Company: Harvey (harvey.ai)
- Announced: 2026-03-18T12:30:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/introducing-harvey-box-integration
- Record: https://forck.live/items/17594-harvey-s-box-integration-your-legal-documents-now-in-harvey
- Subject: Harvey

Harvey announced an integration with Box that lets legal teams access and analyze documents stored in Box directly within Harvey Assistant, Workflow agents, and vaults, without creating local copies. The integration uses OAuth-scoped access so Harvey only touches files users explicitly select, preserving existing governance controls. It is live for current Harvey customers.

### Vercel — 360 billion tokens, 3 million customers, 6 engineers

- Company: Vercel (vercel.com)
- Announced: 2026-03-18T04:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/360-billion-tokens-3-million-customers-6-engineers
- Record: https://forck.live/items/20731-360-billion-tokens-3-million-customers-6-engineers
- Subject: v0 / AI SDK / AI Gateway

Copy link to heading Impact at a glance Durable ships new production agents to customers in a single day AI features and agents serve ~1.1B tokens per day (360B per year) 10x leverage for every engineer, product manager, and designer 3-4x lower infra cost compared to self …

### Krea — Krea Node Agent: Build a Workflow From One Sentence

- Company: Krea (krea.ai)
- Announced: 2026-03-18
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/ai-workflow-agent
- Record: https://forck.live/items/15085-krea-node-agent-build-a-workflow-from-one-sentence
- Subject: Krea 2 / AI creative platform

Krea introduced the Node Agent, a feature within its Nodes product that converts a natural-language description into a runnable creative workflow. The agent reads the existing canvas state, plans the pipeline, validates parameters and connections, estimates compute cost per node, and only executes after user approval. It reuses cached results from upstream nodes when edits are made downstream.

### LG AI Research — [AAAI 2026] AEGIS: An industry anomaly detection system evolving alongside field experts

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-03-18
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=633
- Record: https://forck.live/items/4756-aaai-2026-aegis-an-industry-anomaly-detection-system-evolving-alongside-field
- Subject: EXAONE

LG AI Research introduces AEGIS, an anomaly detection platform that integrates domain expert knowledge through LLM-based explanations, intuitive feedback, and continual learning, presented at AAAI 2026.

### MiniMax — MiniMax M2.7: Early Echoes of Self-Evolution

- Company: MiniMax (minimax.io)
- Announced: 2026-03-18
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.minimax.io/blog/minimax-m27
- Record: https://forck.live/items/3399-minimax-m2-7-early-echoes-of-self-evolution
- Subject: MiniMax models

MiniMax announces M2.7, its first model that participates in its own evolution, with native Agent Teams and self-evolving capabilities, excelling in software engineering, professional work, and entertainment.

### Together AI — Together AI expands fine-tuning service with tool calling, reasoning, and vision support

- Company: Together AI (together.ai)
- Announced: 2026-03-18
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/fine-tuning-update
- Record: https://forck.live/items/2367-together-ai-expands-fine-tuning-service-with-tool-calling-reasoning-and-vision
- Subject: Inference platform

Together AI expands its fine-tuning service with support for tool calling, reasoning, and vision-language model fine-tuning, along with improved training infrastructure for large models.

### Google Research — Google Research at The Check Up: from healthcare innovation to real-world care settings

- Company: Google Research (research.google)
- Announced: 2026-03-17T19:47:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/google-research-at-the-check-up-from-healthcare-innovation-to-real-world-care-settings/
- Record: https://forck.live/items/1356-google-research-at-the-check-up-from-healthcare-innovation-to-real-world-care
- Subject: Research

Google Research at The Check Up event presented multiple AI healthcare innovations including Personal Health Agent (PHA) for personalized health, AI for breast cancer detection and diabetic retinopathy screening, AMIE multi-agent system for clinical collaboration, and Health AI Developer Foundations (HAI-DEF) with MedGemma open-weight models for developers.

### Google Research — Improving breast cancer screening workflows with machine learning

- Company: Google Research (research.google)
- Announced: 2026-03-17T16:57:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/improving-breast-cancer-screening-workflows-with-machine-learning/
- Record: https://forck.live/items/1357-improving-breast-cancer-screening-workflows-with-machine-learning
- Subject: Research

Google Research published two studies in Nature Cancer evaluating an AI-based breast cancer detection system. The first study assessed standalone AI performance and integration feasibility across multiple NHS screening services, showing the AI system achieved higher sensitivity than human readers without compromising specificity, and detected 25% of interval cancers missed in the double-read workflow. The second study compared the AI as a second reader against the original double-read and arbitration process. The results strengthen evidence for AI's potential in supporting breast cancer screening.

### Hugging Face — State of Open Source on Hugging Face: Spring 2026

- Company: Hugging Face (huggingface.co)
- Announced: 2026-03-17T16:37:55+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026
- Record: https://forck.live/items/1533-state-of-open-source-on-hugging-face-spring-2026
- Subject: Platform

Hugging Face published a report analyzing open source AI trends in Spring 2026, covering competition, geographic shifts, and community growth, with data showing China surpassing the US in downloads and independent developers gaining share.

### Google DeepMind — Measuring progress toward AGI: A cognitive framework

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-03-17T16:03:47+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/measuring-progress-toward-agi-a-cognitive-framework/
- Record: https://forck.live/items/1247-measuring-progress-toward-agi-a-cognitive-framework
- Subject: Gemini

Google DeepMind introduces a framework to measure progress toward AGI and launches a Kaggle hackathon to build relevant evaluations.

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

- Company: Vercel (vercel.com)
- Announced: 2026-03-17T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/vercel-open-source-program-winter-2026-cohort
- Record: https://forck.live/items/20732-vercel-open-source-program-winter-2026-cohort
- Subject: v0 / AI SDK / AI Gateway

Vercel announced the Winter 2026 cohort of its Open Source Program, which provides maintainers with resources, credits, and support. The cohort includes 25 projects such as Answer Overflow, UI TripleD, hot-updater, and Screenshot Studio, spanning AI-native apps, developer infrastructure, design systems, and creative tooling.

### H Company — Holotron-12B - High Throughput Computer Use Agent

- Company: H Company (hcompany.ai)
- Announced: 2026-03-17T12:33:39+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/Hcompany/holotron-12b
- Record: https://forck.live/items/1534-holotron-12b-high-throughput-computer-use-agent

H Company releases Holotron-12B, a multimodal computer-use model post-trained from NVIDIA's Nemotron-Nano-2 VL model, featuring a hybrid SSM architecture for high-throughput inference. On the WebVoyager benchmark, it achieves over 2x throughput compared to Holo2-8B on a single H100 GPU.

### OpenAI — OpenAI Japan announces Japan Teen Safety Blueprint to put teen safety first

- Company: OpenAI (openai.com)
- Announced: 2026-03-17T10:00:00+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/japan-teen-safety-blueprint
- Record: https://forck.live/items/228-openai-japan-announces-japan-teen-safety-blueprint-to-put-teen-safety-first
- Subject: GPT / ChatGPT / API

OpenAI Japan announces the Japan Teen Safety Blueprint, which introduces stronger age protections, parental controls, and well-being safeguards for teens using generative AI.

### OpenAI — Introducing GPT-5.4 mini and nano

- Company: OpenAI (openai.com)
- Announced: 2026-03-17T10:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://openai.com/index/introducing-gpt-5-4-mini-and-nano
- Record: https://forck.live/items/227-introducing-gpt-5-4-mini-and-nano
- Subject: GPT / ChatGPT / API

OpenAI introduces GPT-5.4 mini and nano, smaller and faster versions of GPT-5.4 optimized for coding, tool use, multimodal reasoning, and high-volume API and sub-agent workloads.

### Swiss AI Initiative — Apertus for Ticino

- Company: Swiss AI Initiative (swiss-ai.org)
- Announced: 2026-03-17T09:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://apertus-ai.org/articles/2026-03-ticino/
- Record: https://forck.live/items/18377-apertus-for-ticino
- Subject: Apertus

The Canton of Ticino is introducing an in-house AI translation system based on the open-source Apertus model, developed by Artificialy. The fine-tuned Apertus 8b model achieved a benchmark score of 94 percent after fine-tuning, compared to 90 percent before. The solution is hosted in the canton's own data centre to ensure data confidentiality and sovereignty.

### Google DeepMind — Broadening advanced AI education across Africa

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-03-17T08:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.google/company-news/inside-google/around-the-globe/google-africa/ai-research-foundations/
- Record: https://forck.live/items/1319-broadening-advanced-ai-education-across-africa
- Subject: Gemini

Google DeepMind, Google.org, and Google Skills are collaborating to provide a free, high-impact curriculum to train AI researchers and educators in Africa.

### Mistral AI — Introducing Forge

- Company: Mistral AI (mistral.ai)
- Announced: 2026-03-17
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/forge
- Record: https://forck.live/items/3300-introducing-forge
- Subject: Mistral / Le Chat

Mistral AI introduces Forge, a system for enterprises to build frontier-grade AI models using their proprietary knowledge.

### Together AI — Mamba-3

- Company: Together AI (together.ai)
- Announced: 2026-03-17
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.together.ai/blog/mamba-3
- Record: https://forck.live/items/2368-mamba-3
- Subject: Inference platform

Mamba-3 is a new state space model designed with inference efficiency as the primary goal, featuring a more expressive recurrence formula, complex-valued state tracking, and a MIMO variant that boosts accuracy without slowing down decoding. The model outperforms Mamba-2, Gated DeltaNet, and Llama-3.2-1B on prefill+decode latency at the 1.5B scale. The team open-sourced the kernels built using Triton, TileLang, and CuTe DSL.

### OpenAI — Equipping workers with insights about compensation

- Company: OpenAI (openai.com)
- Announced: 2026-03-17
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/equipping-workers-with-insights-about-compensation
- Record: https://forck.live/items/229-equipping-workers-with-insights-about-compensation
- Subject: GPT / ChatGPT / API

OpenAI announces research showing that Americans send nearly 3 million daily messages to ChatGPT asking about compensation and earnings, which helps close the wage information gap.

### Mistral AI — Mistral partners with NVIDIA to accelerate open frontier models

- Company: Mistral AI (mistral.ai)
- Announced: 2026-03-16T20:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/mistral-ai-and-nvidia-partner-to-accelerate-open-frontier-models/
- Record: https://forck.live/items/19883-mistral-partners-with-nvidia-to-accelerate-open-frontier-models
- Subject: Mistral / Le Chat

Mistral AI announced a partnership with NVIDIA to accelerate open frontier models.

### Google Research — Testing LLMs on superconductivity research questions

- Company: Google Research (research.google)
- Announced: 2026-03-16T17:31:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/testing-llms-on-superconductivity-research-questions/
- Record: https://forck.live/items/1358-testing-llms-on-superconductivity-research-questions
- Subject: Research

Google Research published a study testing six LLMs on expert-level questions about high-temperature superconductivity. The top performers were NotebookLM and a custom-built system that used curated references. The paper was published in PNAS.

### Harvey — Announcing Harvey’s Inaugural In-House Customer Advisory Board

- Company: Harvey (harvey.ai)
- Announced: 2026-03-16T14:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harveys-in-house-customer-advisory-board
- Record: https://forck.live/items/17595-announcing-harvey-s-inaugural-in-house-customer-advisory-board
- Subject: Harvey

Harvey announced the formation of its inaugural In-House Customer Advisory Board, a group of General Counsel and Chief Legal Officers from companies including Dentsu, MongoDB, HSBC, NBCUniversal, Koch, and Bridgewater Associates. The board will provide strategic input on Harvey's direction, inform product priorities, and discuss the future of the legal industry with AI. Harvey stated it now works with more than 500 in-house legal teams worldwide.

### Vercel — Meet the 2026 Vercel AI Accelerator Cohort

- Company: Vercel (vercel.com)
- Announced: 2026-03-16T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/2026-vercel-ai-accelerator-cohort
- Record: https://forck.live/items/20733-meet-the-2026-vercel-ai-accelerator-cohort
- Subject: v0 / AI SDK / AI Gateway

Vercel announced the 2026 cohort of its AI Accelerator, selecting 39 early-stage teams from the US, Europe, Asia, and Latin America for a six-week program. Participants receive access to Vercel infrastructure, over $8 million in partner credits from companies including AWS, Anthropic, OpenAI, Cartesia, and ElevenLabs, weekly sessions with founders and technical leaders, and dedicated VC mentors.

### Laminar — Laminar raised $3M to build observability for long-running agents

- Company: Laminar (laminar.sh)
- Announced: 2026-03-16
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://laminar.sh/blog/2026-03-16-laminar-launch
- Record: https://forck.live/items/16852-laminar-raised-3m-to-build-observability-for-long-running-agents
- Subject: flow / agent observability

Laminar announced a $3M seed round led by Atlantic.vc, with participation from Y Combinator, AAL.vc, and angel investors, to build observability for long-running AI agents. The company says existing observability platforms are data collectors that leave users to interpret traces, whereas Laminar is designed to address the failure-analysis and pattern-identification problems of long-running agents in production.

### Mistral AI — Leanstral: Open-Source foundation for trustworthy vibe-coding

- Company: Mistral AI (mistral.ai)
- Announced: 2026-03-16
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/leanstral
- Record: https://forck.live/items/3302-leanstral-open-source-foundation-for-trustworthy-vibe-coding
- Subject: Mistral / Le Chat

Mistral AI announces Leanstral, an open-source foundation for trustworthy vibe-coding.

### Mistral AI — Mistral AI partners with NVIDIA to accelerate open frontier models

- Company: Mistral AI (mistral.ai)
- Announced: 2026-03-16
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/mistral-ai-and-nvidia-partner-to-accelerate-open-frontier-models
- Record: https://forck.live/items/3301-mistral-ai-partners-with-nvidia-to-accelerate-open-frontier-models
- Subject: Mistral / Le Chat

Mistral AI announces a partnership with NVIDIA to accelerate open frontier models.

### Mistral AI — Introducing Mistral Small 4

- Company: Mistral AI (mistral.ai)
- Announced: 2026-03-16
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://mistral.ai/news/mistral-small-4
- Record: https://forck.live/items/3286-introducing-mistral-small-4
- Subject: Mistral / Le Chat

Mistral AI introduces Mistral Small 4, a new model.

### Together AI — Together AI at NVIDIA GTC 2026: Explore our latest innovations across research and products

- Company: Together AI (together.ai)
- Announced: 2026-03-16
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/together-ai-at-nvidia-gtc-2026
- Record: https://forck.live/items/2369-together-ai-at-nvidia-gtc-2026-explore-our-latest-innovations-across-research
- Subject: Inference platform

Together AI announces the availability of NVIDIA Nemotron 3 Super on its platform, a hybrid mixture-of-experts model with 120B parameters and 1M-token context window for multi-agent workflows.

### OpenAI — Why Codex Security Doesn’t Include a SAST Report

- Company: OpenAI (openai.com)
- Announced: 2026-03-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/why-codex-security-doesnt-include-sast
- Record: https://forck.live/items/230-why-codex-security-doesn-t-include-a-sast-report
- Subject: GPT / ChatGPT / API

Codex Security does not use traditional SAST, instead using AI-driven constraint reasoning and validation to reduce false positives.