# forck.live — 9–15 February 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-W07, 9–15 February 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-W07
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Alibaba — Qwen3.5: Towards Native Multimodal Agents

- Company: Alibaba (alibaba.com)
- Announced: 2026-02-15T20:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwen3.5
- Record: https://forck.live/items/4619-qwen3-5-towards-native-multimodal-agents
- Subject: Qwen

Alibaba releases Qwen3.5, an open-weight native vision-language model with 397B total parameters (17B activated), featuring hybrid linear attention and sparse MoE architecture, a 1M token context window, and support for 201 languages.

### Sakana AI — オフィス移転のお知らせ

- Company: Sakana AI (sakana.ai)
- Announced: 2026-02-15T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://sakana.ai/azabudai-hills
- Record: https://forck.live/items/4696-
- Title in English (translated by forck.live from ja): Office Relocation Announcement
- Subject: Sakana models

Sakana AI announced the relocation of its head office to Azabudai Hills Mori JP Tower, citing business expansion and organizational strengthening.

### Sarvam AI — Introducing Sarvam Akshar

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-02-15T07:20:00+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sarvam.ai/blogs/sarvam-akshar/
- Record: https://forck.live/items/18390-introducing-sarvam-akshar
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI released Akshar, a document intelligence workbench built on its Sarvam Vision model. The platform adds visual grounding, layout understanding, and agent-based proofreading to move beyond passive text extraction. It targets last-mile accuracy in complex documents such as historical manuscripts and multi-column layouts.

### Ant Ling — Using Ring 1T with Claude Code via ZenMux

- Company: Ant Ling (ant-ling.com)
- Announced: 2026-02-14T16:00:03+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://developer.ant-ling.com/en/blogs/using-ring-1t-with-claude-code-via-zenmux
- Record: https://forck.live/items/22355-using-ring-1t-with-claude-code-via-zenmux
- Subject: Ling / Ring / Ming

Ant Ling published a guide on using the open-source Ring 1T reasoning model (1 trillion total parameters, 50 billion activated, 128K context) with Anthropic's Claude Code via the ZenMux aggregation platform. The guide explains how to configure Claude Code to route requests through ZenMux using environment variables, and covers ZenMux subscription plans (Pro $20/mo, Max $100/mo, Ultra $200/mo) and pay-as-you-go billing.

### Sarvam AI — Announcing Sarvam Edge

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-02-14T12:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sarvam.ai/blogs/sarvam-edge/
- Record: https://forck.live/items/18391-announcing-sarvam-edge
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI announced Sarvam Edge, an effort to bring AI inference to devices. The initial release includes on-device speech recognition and speech synthesis models supporting 10 Indian languages, with a 74M-parameter ASR model that achieves under 300ms latency and runs entirely locally without cloud dependency.

### MiniMax — Forge: Scalable Agent RL Framework and Algorithm

- Company: MiniMax (minimax.io)
- Announced: 2026-02-14
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.minimax.io/blog/forge-scalable-agent-rl-en-1779896141
- Record: https://forck.live/items/3400-forge-scalable-agent-rl-framework-and-algorithm
- Subject: MiniMax models

Forge, a scalable reinforcement learning framework and algorithm designed to address the trade-off between system throughput, training stability, and agent flexibility for real-world agents.

### ByteDance — Seed 2.0 Official Launch

- Company: ByteDance (bytedance.com)
- Announced: 2026-02-13T16:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://seed.bytedance.com/en/blog/seed-2-0-official-launch
- Record: https://forck.live/items/4667-seed-2-0-official-launch
- Subject: Seed / Doubao

ByteDance announces the Seed2.0 series of large language models, optimized for large-scale production deployment and featuring improved multimodal understanding, instruction following, and inference options. The series includes Pro, Lite, and Mini models, plus a dedicated Code model, and ranks highly on LMSYS Chatbot Arena benchmarks.

### Sarvam AI — Introducing Sarvam Studio

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-02-13T12:00:00+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sarvam.ai/blogs/sarvam-studio/
- Record: https://forck.live/items/18392-introducing-sarvam-studio
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI introduced Sarvam Studio, a platform that combines voice, text, and document workflows into a single workspace for transforming multilingual content while preserving meaning, structure, and intent. It is being rolled out through a controlled early beta program with a small group of production partners across government, education, media, and publishing.

### OpenAI — GPT-5.2 derives a new result in theoretical physics

- Company: OpenAI (openai.com)
- Announced: 2026-02-13T11:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/new-result-theoretical-physics
- Record: https://forck.live/items/270-gpt-5-2-derives-a-new-result-in-theoretical-physics
- Subject: GPT / ChatGPT / API

GPT-5.2 proposed a new formula for a gluon amplitude, which was later formally proved and verified by OpenAI and academic collaborators, as described in a new preprint.

### OpenAI — Introducing Lockdown Mode and Elevated Risk labels in ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2026-02-13T10:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-lockdown-mode-and-elevated-risk-labels-in-chatgpt
- Record: https://forck.live/items/271-introducing-lockdown-mode-and-elevated-risk-labels-in-chatgpt
- Subject: GPT / ChatGPT / API

OpenAI introduces Lockdown Mode and Elevated Risk labels in ChatGPT to help organizations defend against prompt injection and AI-driven data exfiltration.

### OpenAI — Beyond rate limits: scaling access to Codex and Sora

- Company: OpenAI (openai.com)
- Announced: 2026-02-13T09:00:00+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/beyond-rate-limits
- Record: https://forck.live/items/273-beyond-rate-limits-scaling-access-to-codex-and-sora
- Subject: GPT / ChatGPT / API

OpenAI describes how it built a real-time access system using rate limits, usage tracking, and credits to enable continuous access to its products Sora and Codex.

### OpenAI — Scaling social science research

- Company: OpenAI (openai.com)
- Announced: 2026-02-13T09:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/scaling-social-science-research
- Record: https://forck.live/items/272-scaling-social-science-research
- Subject: GPT / ChatGPT / API

OpenAI released GABRIEL, an open-source toolkit that uses GPT to convert qualitative text and images into quantitative data for social science research.

### Tencent — Stabilizing RLVR via Token-level Gradient Diagnosis and Layerwise Clipping

- Company: Tencent (tencent.com)
- Announced: 2026-02-13T08:36:03+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://hunyuan.tencent.com/research/100015
- Record: https://forck.live/items/4581-stabilizing-rlvr-via-token-level-gradient-diagnosis-and-layerwise-clipping
- Subject: Hunyuan

Tencent presents GradLoc, a white-box diagnostic tool for localizing gradient spikes to specific culprit tokens in RLVR training, and proposes LayerClip for layerwise gradient clipping to improve training stability.

### Supabase — Supabase incident on February 12, 2026

- Company: Supabase (supabase.com)
- Announced: 2026-02-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/supabase-incident-on-february-12-2026
- Record: https://forck.live/items/21301-supabase-incident-on-february-12-2026
- Subject: Supabase

Supabase experienced a 3-hour-42-minute outage on February 12, 2026, affecting all services in the us-east-2 region. The cause was a deployment that inadvertently enabled AWS VPC Block Public Access at the regional level, blocking internet gateway traffic. The company has implemented guardrails, access controls, account isolation, and external connectivity probes to prevent recurrence.

### Perceptron — SkillRater: Untangling Capabilities in Multimodal Data

- Company: Perceptron (perceptron.inc)
- Announced: 2026-02-13
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.perceptron.inc/blog/skillrater
- Record: https://forck.live/items/18617-skillrater-untangling-capabilities-in-multimodal-data
- Subject: Perceptron Isaac / Perceptron Mk1

Perceptron introduced SkillRater, a method that decomposes data quality filtering into specialized raters per capability (visual understanding, OCR, STEM) using bilevel meta-learning. The raters show near-orthogonal scoring (mean absolute pairwise Pearson correlation 0.020) and outperform both unfiltered training and monolithic learned filtering on held-out benchmarks at 2B parameters. Raters trained at 1B parameters transfer directly to 2B without retraining.

### LG AI Research — [AAAI 2026] Advances in Multi-Robot Simulation and Coordination for Industrial Automation

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-02-13
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=625
- Record: https://forck.live/items/4760-aaai-2026-advances-in-multi-robot-simulation-and-coordination-for-industrial
- Subject: EXAONE

LG AI Research presented the RAPID platform at the AAAI 2026 Demonstration Program. RAPID is a rapid prototyping platform for industrial automation that enables quick design and analysis of multi-robot coordination in warehouses and factories, addressing the need for fast simulation-based validation and standardization for research.

### Hugging Face — Custom Kernels for All from Codex and Claude

- Company: Hugging Face (huggingface.co)
- Announced: 2026-02-13
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/custom-cuda-kernels-agent-skills
- Record: https://forck.live/items/1547-custom-kernels-for-all-from-codex-and-claude
- Subject: Platform

Hugging Face releases an agent skill that enables coding agents (Claude, Codex) to write production CUDA kernels for transformers and diffusers, including PyTorch bindings and benchmarks.

### Google DeepMind — Gemini 3 Deep Think: Advancing science, research and engineering

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-02-12T16:15:09+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/gemini-3-deep-think-advancing-science-research-and-engineering/
- Record: https://forck.live/items/1254-gemini-3-deep-think-advancing-science-research-and-engineering
- Subject: Gemini

Google DeepMind updated its most specialized reasoning mode to address challenges in science, research, and engineering.

### ByteDance — Deeper Thinking, More Accurate Generation | Introducing Seedream 5.0 Lite

- Company: ByteDance (bytedance.com)
- Announced: 2026-02-12T16:00:00+00:00
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://seed.bytedance.com/en/blog/deeper-thinking-more-accurate-generation-introducing-seedream-5-0-lite
- Record: https://forck.live/items/4668-deeper-thinking-more-accurate-generation-introducing-seedream-5-0-lite
- Subject: Seed / Doubao

ByteDance introduces Seedream 5.0 Lite, an image generation model with improved understanding, reasoning, generation, real-time search, and world knowledge.

### Harvey — Harvey to Open Dallas Office This April

- Company: Harvey (harvey.ai)
- Announced: 2026-02-12T15:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/harvey-to-open-dallas-office-this-april
- Record: https://forck.live/items/17613-harvey-to-open-dallas-office-this-april
- Subject: Harvey

Harvey announced it will open a new office in Dallas, Texas in April, expanding its presence in the legal market. The office will house teams across sales, legal engineering, and customer success to support existing partners including AT&T, KBR, and several law firms.

### OpenAI — Introducing GPT-5.3-Codex-Spark

- Company: OpenAI (openai.com)
- Announced: 2026-02-12T10:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://openai.com/index/introducing-gpt-5-3-codex-spark
- Record: https://forck.live/items/274-introducing-gpt-5-3-codex-spark
- Subject: GPT / ChatGPT / API

OpenAI announced GPT-5.3-Codex-Spark, its first real-time coding model with 15x faster generation and a 128k context window, available in research preview for ChatGPT Pro users.

### Odyssey — Odyssey Announces Investment from NVentures & Samsung Next

- Company: Odyssey (odyssey.systems)
- Announced: 2026-02-12
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://odyssey.systems/investment-from-nvidia-and-samsung
- Record: https://forck.live/items/18601-odyssey-announces-investment-from-nventures-samsung-next
- Subject: Odyssey world models

Odyssey announced an investment from NVentures (NVIDIA's venture capital arm) and Samsung Next to accelerate research towards a general-purpose world simulator. The company's latest model, Odyssey-2 Pro, is described as a breakthrough general-purpose world model available for integration into applications such as robotics, gaming, education, and defense.

### Kyutai — Hibiki-Zero: Simultaneous Speech-to-Speech Translation Without Aligned Data

- Company: Kyutai (kyutai.org)
- Announced: 2026-02-12
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://kyutai.org/blog/2026-02-12-hibiki-zero
- Record: https://forck.live/items/18348-hibiki-zero-simultaneous-speech-to-speech-translation-without-aligned-data
- Subject: Moshi / Pocket TTS

Real-time speech translation from four languages.

### Z.ai — GLM-5

- Company: Z.ai (z.ai)
- Announced: 2026-02-12
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://docs.z.ai/release-notes/new-released#2026-02-12
- Record: https://forck.live/items/4451-glm-5
- Subject: GLM

GLM-5 is a new model from Z.ai designed for complex system engineering and long-range Agent tasks, with deep-reasoning performance in backend architecture, algorithms, and bug fixing. It benchmarks against Claude Opus 4.5 and integrates DeepSeek Sparse Attention for token efficiency.

### MiniMax — MiniMax M2.5: Built for Real-World Productivity.

- Company: MiniMax (minimax.io)
- Announced: 2026-02-12
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.minimax.io/blog/minimax-m25
- Record: https://forck.live/items/3401-minimax-m2-5-built-for-real-world-productivity
- Subject: MiniMax models

MiniMax M2.5 is built for real-world productivity.

### Cursor — Long-running Agents in Research Preview

- Company: Cursor (cursor.com)
- Announced: 2026-02-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/02-12-26
- Record: https://forck.live/items/2634-long-running-agents-in-research-preview
- Subject: Cursor

Cursor announces that its agent can now work autonomously over longer horizons to complete larger, more complex tasks without human intervention, and that this long-running agent capability is available for Ultra, Teams, and Enterprise plans.

### Together AI — Introducing Dedicated Container Inference: Delivering 2.6x faster inference for custom AI models

- Company: Together AI (together.ai)
- Announced: 2026-02-12
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/dedicated-container-inference
- Record: https://forck.live/items/2380-introducing-dedicated-container-inference-delivering-2-6x-faster-inference-for
- Subject: Inference platform

Together AI announces Dedicated Container Inference, a service for deploying custom generative media models with production-grade orchestration, autoscaling, and monitoring. Customers have seen 1.4x to 2.6x inference speedups.

### Hugging Face — OpenEnv in Practice: Evaluating Tool-Using Agents in Real-World Environments

- Company: Hugging Face (huggingface.co)
- Announced: 2026-02-12
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/openenv-turing
- Record: https://forck.live/items/1548-openenv-in-practice-evaluating-tool-using-agents-in-real-world-environments
- Subject: Platform

OpenEnv, an open-source framework from Meta and Hugging Face for evaluating AI agents in real-world environments rather than simulations. It describes a production-grade calendar management environment (Calendar Gym) built by Turing to test tool-using agents under realistic constraints like access control and multi-step reasoning. The post reports that agents often perform well on individual actions but struggle with multi-step reasoning and reliability in longer workflows.

### Stability AI — Stability AI Joins the Tech Coalition

- Company: Stability AI (stability.ai)
- Announced: 2026-02-11T20:28:23+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://stability.ai/news-updates/stability-ai-joins-the-tech-coalition
- Record: https://forck.live/items/17477-stability-ai-joins-the-tech-coalition
- Subject: Stable Diffusion

Stability AI has joined the Tech Coalition, a global alliance of technology companies working to combat online child sexual exploitation and abuse. The company previously participated in the Tech Coalition's Pathways program in 2025 and is now a full member, reflecting the maturity of its integrity and safety program.

### Harvey — Building a New File Ingestion System to Scale Firm Knowledge

- Company: Harvey (harvey.ai)
- Announced: 2026-02-11T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/building-new-file-ingestion-system-to-scale-firm-knowledge
- Record: https://forck.live/items/17614-building-a-new-file-ingestion-system-to-scale-firm-knowledge
- Subject: Harvey

Harvey built a new asynchronous file ingestion system using Temporal to handle hundreds of thousands of files from document management systems. The system adds folder uploads and one-way sync, so changes in the DMS are automatically detected and synced. The architecture uses granular Temporal activities for crawling and downloading, with automatic retries and checkpointing to resume from failures.

### ByteDance — Seedance 2.0 Official Launch

- Company: ByteDance (bytedance.com)
- Announced: 2026-02-11T16:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://seed.bytedance.com/en/blog/seedance-2-0-official-launch
- Record: https://forck.live/items/4669-seedance-2-0-official-launch
- Subject: Seed / Doubao

ByteDance released Seedance 2.0, a video generation model with multimodal input (text, image, audio, video) and enhanced controllability and motion realism.

### Sarvam AI — Introducing Saaras V3

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-02-11T12:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.sarvam.ai/blogs/asr/
- Record: https://forck.live/items/18393-introducing-saaras-v3
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI released Saaras V3, a speech recognition model supporting 22 Indian languages and English with native streaming for low-latency transcription. On the IndicVoices benchmark, it achieves a word error rate of ~19%, improving over the previous version's ~22% WER. The model is trained on over 1 million hours of curated multilingual audio and maintains accuracy in both batch and real-time configurations.

### ElevenLabs — Klarna reduces Time to Resolution by 10X with ElevenAgents

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-02-11T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://elevenlabs.io/blog/klarna
- Record: https://forck.live/items/3365-klarna-reduces-time-to-resolution-by-10x-with-elevenagents
- Subject: Voice AI

Klarna achieved a 10x reduction in Time to Resolution using ElevenAgents.

### ElevenLabs — Introducing ElevenLabs for Government

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-02-11T12:00:00+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://elevenlabs.io/blog/introducing-elevenlabs-for-government
- Record: https://forck.live/items/3363-introducing-elevenlabs-for-government
- Subject: Voice AI

ElevenLabs is launching a new product called ElevenLabs for Government.

### Google Research — Scheduling in a changing world: Maximizing throughput with time-varying capacity

- Company: Google Research (research.google)
- Announced: 2026-02-11T10:34:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/scheduling-in-a-changing-world-maximizing-throughput-with-time-varying-capacity/
- Record: https://forck.live/items/1366-scheduling-in-a-changing-world-maximizing-throughput-with-time-varying-capacity
- Subject: Research

Google Research introduces new provably effective algorithms for scheduling non-preemptive jobs on cloud infrastructure with time-varying capacity, providing constant-factor approximation guarantees for throughput maximization in both offline and online settings.

### OpenAI — Harness engineering: leveraging Codex in an agent-first world

- Company: OpenAI (openai.com)
- Announced: 2026-02-11T09:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/harness-engineering
- Record: https://forck.live/items/275-harness-engineering-leveraging-codex-in-an-agent-first-world
- Subject: GPT / ChatGPT / API

No announcement content.

### Gradium — Optimizing Quality vs. Latency in Real-Time Text-to-Speech AI Models

- Company: Gradium (gradium.ai)
- Announced: 2026-02-11
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://gradium.ai/blog/optimizing-quality-vs-latency
- Record: https://forck.live/items/18327-optimizing-quality-vs-latency-in-real-time-text-to-speech-ai-models
- Subject: Gradium TTS / Phonon

Gradium published a technical guide covering latency and quality optimization for its streaming text-to-speech models. The post explains how the Delayed Streams Modeling architecture and techniques such as CUDA graphs and variable codebook counts allow users to trade audio quality for lower time-to-first-audio and higher real-time factor on NVIDIA GPUs.

### Luma — Luma AI To Open Riyadh Office To Accelerate Humain Create And Partner With Publicis Groupe

- Company: Luma (lumalabs.ai)
- Announced: 2026-02-11
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lumalabs.ai/news/luma-ai-to-open-riyadh-office-to-accelerate-humain-create-and-partner-with-publicis-groupe
- Record: https://forck.live/items/7971-luma-ai-to-open-riyadh-office-to-accelerate-humain-create-and-partner-with
- Subject: Ray / Dream Machine

Luma AI announced it will open a Riyadh office to support its partnership with HUMAIN and the HUMAIN Create initiative, and was named the preferred generative AI technology partner for Publicis Groupe Middle East across the MENA region.

### Runway — New Funding to Scale World Simulation

- Company: Runway (runway.com)
- Announced: 2026-02-10T23:23:58.290000+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/runway-series-e-funding
- Record: https://forck.live/items/7897-new-funding-to-scale-world-simulation
- Subject: Gen / Aleph

Runway announced $315 million in Series E funding, led by General Atlantic with participation from NVIDIA, Adobe Ventures, AllianceBernstein, AMD Ventures, Fidelity Management & Research Company, Mirae Asset, Emphatic Capital, Felicis and Premji Invest, to pre-train the next generation of world models and bring them to new products and industries.

### Google Research — Beyond one-on-one: Authoring, simulating, and testing dynamic human-AI group conversations

- Company: Google Research (research.google)
- Announced: 2026-02-10T18:30:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/beyond-one-on-one-authoring-simulating-and-testing-dynamic-human-ai-group-conversations/
- Record: https://forck.live/items/1367-beyond-one-on-one-authoring-simulating-and-testing-dynamic-human-ai-group
- Subject: Research

Google Research introduces DialogLab, an open-source prototyping framework for authoring, simulating, and testing dynamic human-AI group conversations, presented at ACM UIST 2025.

### Supabase — Hydra joins Supabase

- Company: Supabase (supabase.com)
- Announced: 2026-02-10T17:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/hydra-joins-supabase
- Record: https://forck.live/items/21302-hydra-joins-supabase
- Subject: Supabase

Supabase announced that Joe, co-creator of Hydra, is joining the team to lead the development of Supabase Warehouse, an open data warehouse architecture. Hydra co-developed pg_duckdb, an MIT-licensed Postgres extension that accelerates analytics queries by over 600x. Supabase will maintain Hydra's existing repositories and continue contributing to pg_duckdb.

### fal — Instruction-level control with Inline Elementwise ASM in Triton

- Company: fal (fal.ai)
- Announced: 2026-02-10T16:42:02+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.fal.ai/instruction-level-control-with-inline-elementwise-asm-in-triton/
- Record: https://forck.live/items/8005-instruction-level-control-with-inline-elementwise-asm-in-triton
- Subject: fal platform

Triton provides an inline_asm_elementwise function that allows injecting elementwise PTX instructions into Triton kernels. The post demonstrates using the rcp.approx instruction to compute fast approximate reciprocals. The compilation pipeline lowers Triton kernels through MLIR and LLVM IR to PTX and then to CUBIN.

### Harvey — How Legal Teams are Working Better With 25,000+ Workflows

- Company: Harvey (harvey.ai)
- Announced: 2026-02-10T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/25000-custom-workflows
- Record: https://forck.live/items/17615-how-legal-teams-are-working-better-with-25-000-workflows
- Subject: Harvey

Harvey reports that its customers have created over 25,000 custom workflows, which are multi-step AI agents tailored to legal tasks. The company cites examples of law firms saving up to five hours per week or 75% of diligence review time, and outlines planned investments in workflow discovery, natural-language editing, and analytics.

### Sarvam AI — Introducing Sarvam Arya

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-02-10T12:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sarvam.ai/blogs/introducing-sarvam-arya/
- Record: https://forck.live/items/18394-introducing-sarvam-arya
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI introduced Arya, an agent orchestration stack designed for production-grade agents. The stack provides eight flat, composable primitives (LLM, Agent, MCP, Node, Ledger, Task Graph, Code Interpreter, Artefact) that combine to build reliable multi-agent systems. Arya offers four guarantees: primitives that compose, state that survives crashes, controlled dynamism, and declarative authoring.

### Alibaba — Qwen-Image-2.0: Professional infographics, exquisite photorealism

- Company: Alibaba (alibaba.com)
- Announced: 2026-02-10T05:08:30+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwen-image-2.0
- Record: https://forck.live/items/4601-qwen-image-2-0-professional-infographics-exquisite-photorealism
- Subject: Qwen

Alibaba launched Qwen-Image-2.0, a next-generation image generation model that supports 1k-token instructions for professional infographics, 2K resolution for realistic scenes, unified generation and editing, and a lighter architecture for faster inference.

### Isomorphic Labs — The Isomorphic Labs Drug Design Engine unlocks a new frontier beyond AlphaFold

- Company: Isomorphic Labs (isomorphiclabs.com)
- Announced: 2026-02-10
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier
- Record: https://forck.live/items/19144-the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier-beyond-alphafold
- Subject: Drug Design Engine / AlphaFold

Isomorphic Labs introduced the Isomorphic Labs Drug Design Engine (IsoDDE), a unified computational drug-design system that demonstrates improved predictive accuracy over AlphaFold 3. On a protein-ligand structure prediction benchmark, IsoDDE more than doubles AlphaFold 3's accuracy on the most difficult systems. It also outperforms AlphaFold 3 by 2.3x and Boltz-2 by 19.8x on a challenging antibody-antigen test set, and surpasses all deep-learning methods on three binding affinity prediction benchmarks.

### Krea — Prompt-to-Workflow

- Company: Krea (krea.ai)
- Announced: 2026-02-10
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/prompt-to-workflow
- Record: https://forck.live/items/15091-prompt-to-workflow
- Subject: Krea 2 / AI creative platform

Krea introduced Prompt-to-Workflow, a feature within Krea Nodes that generates a complete multi-step workflow graph from a single sentence describing the desired output. Users can then adjust parameters, swap models, or run the generated pipeline as-is, accelerating tasks such as batch image generation, style transfer, and multi-step editing chains.

### Amp — Amp Free Is Full (For Now)

- Company: Amp (ampcode.com)
- Announced: 2026-02-10
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/amp-free-is-full-for-now
- Record: https://forck.live/items/7757-amp-free-is-full-for-now
- Subject: Amp

Amp is closing admission to its free tier (Amp Free) for new users; existing users retain $10/day in free usage supported by ads. The company cites needing to focus on building the next version of Amp.

### Cognition — Closing the Agent Loop: Devin Autofixes Review Comments

- Company: Cognition (cognition.com)
- Announced: 2026-02-10
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/closing-the-agent-loop-devin-autofixes-review-comments
- Record: https://forck.live/items/7680-closing-the-agent-loop-devin-autofixes-review-comments
- Subject: Devin / SWE

Cognition added an autofix feature to Devin that automatically resolves review comments from GitHub bots, including linters, CI pipelines, and security scanners, and feeds the fixes back into the pull request.

### Google Research — How AI trained on birds is surfacing underwater mysteries

- Company: Google Research (research.google)
- Announced: 2026-02-09T18:38:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/how-ai-trained-on-birds-is-surfacing-underwater-mysteries/
- Record: https://forck.live/items/1368-how-ai-trained-on-birds-is-surfacing-underwater-mysteries
- Subject: Research

Google Research and Google DeepMind present a paper showing that Perch 2.0, a bioacoustics foundation model trained on birds and terrestrial animals, can be used for transfer learning to classify whale vocalizations, and they release a Colab tutorial.

### Google DeepMind — Accelerating Mathematical and Scientific Discovery with Gemini Deep Think

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-02-09T16:12:06+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/accelerating-mathematical-and-scientific-discovery-with-gemini-deep-think/
- Record: https://forck.live/items/1255-accelerating-mathematical-and-scientific-discovery-with-gemini-deep-think
- Subject: Gemini

Research papers indicate that Deep Think is having a growing impact across various fields.

### Harvey — Expanding BigLaw Bench

- Company: Harvey (harvey.ai)
- Announced: 2026-02-09T15:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/expanding-big-law-bench
- Record: https://forck.live/items/17616-expanding-biglaw-bench
- Subject: Harvey

Harvey announced plans to expand its BigLaw Bench (BLB) benchmark, increasing the number of data points more than fivefold to cover global jurisdictions, core legal practice areas, and complex legal research tasks. The company will publish methodological deep dives and updated BLB scores for all major models in the coming months.

### Vercel — Anyone can build agents, but it takes a platform to run them

- Company: Vercel (vercel.com)
- Announced: 2026-02-09T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/anyone-can-build-agents-but-it-takes-a-platform-to-run-them
- Record: https://forck.live/items/20751-anyone-can-build-agents-but-it-takes-a-platform-to-run-them
- Subject: v0 / AI SDK / AI Gateway

Vercel describes how its agent orchestration platform provides infrastructure primitives for building and running internal and customer-facing AI agents, citing its own internal data agent d0 as an example. The post argues that production deployment of agents remains difficult and that a robust platform is needed to handle security, observability, reliability, and cost optimization.

### Vercel — How we built AEO tracking for coding agents

- Company: Vercel (vercel.com)
- Announced: 2026-02-09T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-we-built-aeo-tracking-for-coding-agents
- Record: https://forck.live/items/20750-how-we-built-aeo-tracking-for-coding-agents
- Subject: v0 / AI SDK / AI Gateway

Vercel describes how it built an AI Engine Optimization (AEO) tracking system for coding agents, using Vercel Sandbox to run agents like Claude Code and OpenAI Codex in ephemeral MicroVMs and routing their LLM calls through AI Gateway for unified logging and cost tracking.

### Sarvam AI — Sarvam Announces Sovereign AI Partnerships with Indian States

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-02-09T12:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sarvam.ai/blogs/partnerships-with-indian-states/
- Record: https://forck.live/items/18395-sarvam-announces-sovereign-ai-partnerships-with-indian-states
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI announced strategic partnerships with the Governments of Odisha and Tamil Nadu to build sovereign AI capacity. The collaboration includes a 50MW AI-optimized facility in Odisha and Digital Sangam, India's first Sovereign AI Research Park with IIT Madras in Tamil Nadu, aimed at developing compute infrastructure, sovereign models, and institutional capacity for AI adoption in public services.

### OpenAI — Bringing ChatGPT to GenAI.mil

- Company: OpenAI (openai.com)
- Announced: 2026-02-09T11:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/bringing-chatgpt-to-genaimil
- Record: https://forck.live/items/276-bringing-chatgpt-to-genai-mil
- Subject: GPT / ChatGPT / API

OpenAI announces deployment of ChatGPT on GenAI.mil for U.S. defense teams.

### Swiss AI Initiative — Open Source LLM Builders

- Company: Swiss AI Initiative (swiss-ai.org)
- Announced: 2026-02-09T09:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://apertus-ai.org/articles/2026-02-builder-summit/
- Record: https://forck.live/items/18380-open-source-llm-builders
- Subject: Apertus

The second Open-Source LLM Builders Summit in Lausanne concluded with a plenary discussion synthesizing breakout session insights on global collaboration for open LLMs. The summit surfaced challenges including accelerating knowledge exchange beyond traditional academic channels, moving from publishing models to building deployable systems, navigating regulatory fragmentation, and rethinking team and incentive structures. Participants identified the need for better documentation of training data, sharing of full pipeline practices, and coordinated advocacy with policymakers to enable scalable open-source LLM development.

### Gradium — Building Voice Agents From the Ground Up: The Gradium Startup Program

- Company: Gradium (gradium.ai)
- Announced: 2026-02-09
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://gradium.ai/blog/gradium-startup-program
- Record: https://forck.live/items/18328-building-voice-agents-from-the-ground-up-the-gradium-startup-program
- Subject: Gradium TTS / Phonon

Gradium launched a Startup Program that provides eligible voice-AI startups with six months of its M Plan for free, including 9 million monthly credits, 10 concurrent requests, pro voice cloning, and commercial rights. The program targets seed- or Series A-funded teams with fewer than 25 employees that have voice as a core part of their product and a clear monetization strategy. Accepted founders also receive direct engineering support and previews of upcoming models.

### Moonshot AI — Agent Swarm

- Company: Moonshot AI (moonshot.ai)
- Announced: 2026-02-09
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.kimi.ai/blog/agent-swarm
- Record: https://forck.live/items/4624-agent-swarm
- Title in English (translated by forck.live from zh): Kimi Introduces Agent Swarm: Let 100 AI Agents Work for You
- Subject: Kimi

Moonshot AI announced Agent Swarm, a new system where multiple AI agents can work in parallel, self-organize into teams with roles, and execute tasks like research or report writing faster than sequential models. It is built on Kimi K2.5 and can deploy up to 100 sub-agents and over 1,500 tool calls, producing results 4.5x faster than sequential execution.

### Hugging Face — Transformers.js v4: Now Available on NPM!

- Company: Hugging Face (huggingface.co)
- Announced: 2026-02-09
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/transformersjs-v4
- Record: https://forck.live/items/1549-transformers-js-v4-now-available-on-npm
- Subject: Platform

Hugging Face released Transformers.js v4, a new major version of the JavaScript library for running AI models, featuring a new WebGPU runtime, new models, repository restructuring, and performance improvements.