Skip to content
Source · Daily Brief

AI Morning Brief — 10 July 2026

OpenAI released GPT-5.6 to the public on July 9, pairing the three-tier model family with ChatGPT Work, a new agent capable of running multi-hour workflows across enterprise apps. Within hours, Meta countered with Muse Spark 1.1 on a new Meta Model API priced at a fraction of any competing frontier model.

Top stories

  • GPT-5.6 launches in three tiers alongside ChatGPT Work. After a limited government preview, OpenAI received US regulatory clearance and shipped Sol ($5/$30 per 1M tokens), Terra ($2.50/$15), and Luna ($1/$6). ChatGPT Work, powered by Codex and GPT-5.6, handles complete workflows across Slack, Google Drive, and Salesforce without ongoing user input; Atlas, OpenAI’s standalone AI browser, is simultaneously being shut down. GPT-5.6 Sol and Terra are now also available in GitHub Copilot and are the designated preferred model in Microsoft 365 Copilot. via The Verge
  • Meta ships Muse Spark 1.1 and opens the Meta Model API. Mark Zuckerberg announced Muse Spark 1.1 — an agentic and coding model Meta claims rivals GPT-5.5 and Claude Opus 4.8 — available now at roughly $4.25 per million output tokens, or about one-tenth the cost of competing tier-one models. The Meta Model API is in public preview; an open-source variant of Muse Spark is also reportedly in development. via TechCrunch
  • Grok 4.5 benchmark claims accumulate a day after launch. Third-party evaluations cited by xAI place Grok 4.5 at the top of AutomationBench-AA, Terminal-Bench v2, the Harvey Legal Agent Benchmark, and SWE Marathon. A 1M-token context window is expected as soon as next week, and a 2T-parameter successor model is reportedly finishing training this month. via Bens Bites
  • Fidji Simo departs OpenAI. OpenAI’s No. 2 executive is leaving her full-time role and transitioning to a part-time advisor role after a neuroimmune condition extended a medical leave first announced in April. The departure removes a key operational leader as the company prepares for a possible IPO. via TechCrunch
  • The New York Times accuses OpenAI of hiding training data evidence. In escalating copyright litigation, the Times alleges OpenAI falsely claimed an inability to search training data and concealed billions of ChatGPT logs that could identify copyrighted journalism in model outputs; the paper is seeking sanctions. via Ars Technica
  • Ollama closes a $65M Series B. The open-source local-model runtime has grown to nearly 9M active users, 176,000 GitHub stars, and 67,000 integrations across 85% of Fortune 500 companies. Theory Ventures led the round, citing Ollama’s position as the default path for developers running open models. via TechCrunch

Who shipped

OpenAI was the day’s dominant actor: GPT-5.6 in three tiers, ChatGPT Work, a new voice model as GPT-Live, and simultaneous distribution into GitHub Copilot and Microsoft 365 Copilot. Meta followed within the same day with Muse Spark 1.1 and the Meta Model API, its first commercial API offering. xAI‘s day-old Grok 4.5 continued accumulating benchmark results and third-party coverage. Google DeepMind moved AlphaEvolve, its Gemini-powered evolutionary coding agent, to general availability on Google Cloud. Anthropic announced former Federal Reserve Chair Ben Bernanke’s appointment to its Long-Term Benefit Trust, and its Jacobian lens interpretability research drew wide coverage in the technical press.

Open-source pulse

Ollama‘s $65M Series B underscored how embedded it has become as the default local-inference runtime. Netflix released video datasets and models on Hugging Face, drawing attention from the open-source community. Multiple llama.cpp builds (b9941–b9946) landed overnight with Hexagon tile optimisations and ROCm improvements.

Money, infra & hardware

Beyond Ollama’s Series B, Paris-based AI voice startup Gradium announced a $100M seed round backed by Nvidia. Agent startup Lyzr disclosed a $100M fundraise run end-to-end by its own AI agent. Meta confirmed its custom AI chips enter production in September. Microsoft‘s 2026 sustainability report disclosed a 25% rise in carbon emissions in 2025, attributed largely to data-centre expansion for AI workloads.

Quiet corners

DeepSeek generated substantial community discussion — users demonstrated V4 Flash running on a single RTX 6000 Pro — but the lab made no announcements. Alibaba/Qwen and Cohere were also absent from the headline cycle.

By the numbers

  • 310+ stories in 24 h across 30+ sources
  • Most-mentioned model: GPT (GPT-5.6)
  • Most-mentioned lab: OpenAI
  • Notable absences: DeepSeek, Alibaba/Qwen, Cohere

Compiled by AI Feed’s editor from all 310+ headlines published on 10 July 2026.