Models — page 4
OpenAI clears GPT-5.6 for public rollout, launches ChatGPT Work agent
The GPT-5.6 suite — Sol, Terra, and Luna — powers a new agent that pulls context from Slack, Gmail, and Google Drive.
Meta starts production of MTIA AI chips in September to cut Nvidia dependence
The social giant is spending up to $145B this year on AI compute and wants its own silicon to blunt GPU costs.
Meta opens Muse Spark 1.1 to developers via new Meta Model API
The upgraded model targets agentic coding and multimodal reasoning, arriving days after the controversial Muse Image launch.
OpenAI updates ChatGPT voice mode to interrupt users less often
GPT-Live-1 replaces the older turn-based voice model with full-duplex audio that can listen while it speaks.
xAI releases Grok 4.5, priced at less than half of Claude Opus 4.7
Elon Musk calls the new model Opus-class but faster and cheaper, at $2 per million input tokens and $6 per million output.
OpenAI ships GPT-Live-1, a full-duplex voice model that replaces Advanced Voice Mode
The new model listens and speaks simultaneously, routes to GPT-5.5 for reasoning, and now serves 150M weekly voice users.
Nvidia Nemotron 3 Ultra hits closed-model parity at 10x lower cost on LangChain
LangChain tuned its Deep Agents harness for Nemotron 3 Ultra, matching top closed models on business tasks without retraining.
ZML launches free LLMD inference server across Nvidia, AMD, Google TPU and Apple chips
The Paris startup, backed by $20M and Yann LeCun, wants to break silicon lock-in as inference costs climb.
Meta's Muse Image lets Instagram users pull each other into AI photos
Superintelligence Labs ships its first image model, replacing Llama in Meta AI and powering 30 new Instagram Stories effects.
DeepSeek plans its own inference chips to cut Nvidia and Huawei reliance
The Chinese LLM developer has spent about a year on a silicon project targeting data center inference, per Reuters.
Anthropic frees Claude Cowork from the desktop with always-on mobile agent
Cowork now runs tasks overnight without an open laptop, arriving first to Max plan subscribers at $100 a month.
Nvidia's Vera CPU targets agentic AI with 50% IPC gain over Grace
The 88-core Arm chip aims to keep GPUs fed as agents run in loops; Perplexity ran a coding workflow 1.5x faster than x86.
Nvidia and Hugging Face push Isaac GR00T 1.7 into LeRobot for open robotics
The integration connects 3M robotics developers to 16M AI builders, with Cosmos 3 world models coming next to Hugging Face's open library.
Fable tops KernelBench-Mega as AI agents quadruple on real freelance work
A new GPU kernel record, a 4x jump on the Remote Labor Index in eight months, and OSWORLD 2.0 raise the ceiling on what AI agents can do.
Springboards launches Flint, an LLM built to escape the chatbot groupthink rut
Ask ChatGPT, Claude, or Gemini for a random number between 1 and 10 and you'll almost always get 7. Flint answered 3.7916.
Anthropic ships Claude Sonnet 5 at $2 per million input tokens
Sonnet 5 hits 63.2% on agentic coding, close to Opus 4.8's 69.2%, at a fraction of the price.
NVIDIA's inference software stack cuts DeepSeek V4 token costs 5x in one month
Baseten, Cognition, Together AI and Cursor are riding compounding software gains on Blackwell GPUs as inference economics shift to cost per token.
Google ships Nano Banana 2 Lite and Gemini Omni Flash to developers
DeepMind's fastest image model generates in 4 seconds at $0.034 per 1K images; Omni Flash matches Veo 3.1 Fast at $0.10 per second of video.
NVIDIA ships Omniverse skills to train vision AI agents on synthetic defect data
Roboflow hit 95% average precision on Corning fiber defects using just 8 real images plus synthetic data from NVIDIA's new skill.
Flexion Robotics trains humanoids to run office errands autonomously
The Swiss startup, founded by ex-Nvidia researchers, stacks reinforcement learning across every layer to make humanoids work without a human puppeteer.
China's Z.ai claims GLM-5.2 matches Anthropic's Mythos on bug-finding
The open-weight model lags on general tasks but reportedly closes the gap on cybersecurity — the exact capability US export curbs were meant to contain.
Sakana and 360 ship Mythos rivals as US ban on Anthropic exports holds
Tokyo's Sakana AI launched Fugu and China's 360 unveiled Tulongfeng within days of each other, two weeks into Washington's block on Anthropic's frontier models.
OpenAI joins Google, Apple, and SpaceX in building its own AI chips
OpenAI's Jalapeño inference chip, built with Broadcom, marks the latest move by frontier AI buyers to hedge against Nvidia dependence.
OpenAI ships GPT-5.6 in three tiers, undercuts Claude on price
Sol, Terra, and Luna launch under a White House-monitored preview, with Sol priced at half of Anthropic's Claude Fable 5.
Stay ahead of everyone in AI.
The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.
The briefing read inside teams at