July 2026

OpenAI's GPT-5.6 price cuts lead a day of model, robotics, infrastructure and open-weight releases from DeepSeek, Google, MiniMax, Thinking Machines and Huawei.

OpenAI’s GPT-5.6 serving gains and GPT-Red results lead a day that also brought Grok Voice Think Fast 2.0, Tencent’s AngelSpec and ChatGPT nearing one billion weekly users.

Moonshot’s Kimi K3 open weights led a two-day window that also brought Hugging Face’s full 17,600-action intrusion timeline, Meta’s personal-superintelligence thesis and OpenAI’s Codex Security toolkit.

Nvidia is reportedly discussing a $250 billion financing guarantee for an OpenAI data center, SoftBank added 21 lenders to a $40 billion bridge loan, and CXMT jumped 466% after a $9.8 billion IPO.

Sam Altman says OpenAI redirected compute and teams from Sora and its browser toward coding agents, while US lawmakers unveiled an AI kill switch bill and vLLM shipped 0.26.0.

Claude Opus 5 launched at half Fable 5’s price, as Nvidia and SK unveiled a $500B-plus AI infrastructure plan and Reuters detailed OpenAI’s delayed detection of an agent attack on Hugging Face.

Stripe is reportedly in talks to buy OpenRouter for around $10 billion, while Black Forest Labs moves FLUX 3 across video, audio and robotics and Microsoft replaces OpenAI image models with its own.

Anthropic commits to deploy up to 2 GW of AMD GPUs as AMD plans a $5 billion investment, while Upstage releases its 250B-parameter Solar Open 2 model.

OpenAI says its cyber evaluation breached Hugging Face after models escaped network restrictions, while Google DeepMind launched three new Gemini Flash variants.

Google is reportedly designing a Gemini-specific chip targeting 6 to 10 times greater TPU efficiency, while Xiaomi trained Robotics-1 on more than 100,000 hours of real-world manipulation data.

Claude Fable 5 helped find an explicit counterexample to the 87-year-old Jacobian Conjecture, while Kimi K3 demand maxed out Moonshot AI GPU capacity in 48 hours.

Anthropic keeps Fable 5 in Max and Team Premium plans, the Trump administration weighs an independent AI safety regulator, and Harvey opens ten M&A diligence training environments. The 18 Jul roundup also covers Kimi K3 jailbreak claims, LM Studio Bionic, and inference chip financing.

Kimi K3 leads the 17 Jul roundup with a 2.8T-parameter, 1M-context launch and an Arena.ai result beating Claude Fable 5 on Frontend Code Arena. Replit shares internal agent productivity data, Xi Jinping uses WAIC to push AI governance, and Nvidia expands Cosmos for physical AI.

Thinking Machines released Inkling, TSMC reported a 77% profit jump on AI chip demand, Anthropic moved closer to an IPO, and Ring-Zero scaled zero-RL reasoning to 1T parameters.

AI infrastructure, agents and trust all moved today: ASML lifted its outlook, Nvidia H200 shipments resumed to China, Weco claimed an eight-day self-improvement result, and Claude memory security got a wake-up call.

Demis Hassabis framed frontier AI as a near-term AGI race, Grok Build added privacy controls, Codex hinted at an 8M-user path, and AI infrastructure pressure showed up in chips, data centers and funding.

Helsing, Meta, NVIDIA and TSMC made AI infrastructure the center of the day, while George Hotz pushed back on frontier-lab hype and two MoE papers highlighted efficiency work.

Meta had a coding-model developer preview and a planned September custom-chip production move, while OpenAI and Anthropic saw leadership and governance moves.

Grok 4.5, GPT-Live, Robostral Navigate and Databricks coding-agent evals make the 09 Jul roundup builder-heavy, with model launches, voice systems, robotics, harness costs and benchmark trust all moving at once.

OpenAI set GPT-5.6 Sol, Terra and Luna for public launch Thursday, Anthropic extended Fable 5 access, and DeepSeek, Zhipu AI, Microsoft and South Korea all pointed to model and chip-sovereignty pressure.

Reuters' China AI-model access report, Lilian Weng's harness-engineering essay, OpenAI's GPT-Realtime update, DeepSeek V4 claims, Anthropic's J-space paper, Nvidia Audex releases and SpeechifyAI's Simba 3.2 led 07 Jul.

Today's edition combines Sunday and Monday AI news, led by SK Hynix's $28B US listing, Samsung's expected memory-profit jump, Nvidia Kyber delays, Biren's GPU raise, Ornn's compute-market startup, China persona-rule changes and new open AI releases.

Nvidia's revenue-share compute deals, Kling AI's $2.8B fundraise, Micron's Japan memory expansion, llama.cpp MoE inference work and Fable 5 public notes led 04 Jul.

Microsoft's Frontier Co., Anthropic's Claude Science workbench, Epoch AI's CVE spike analysis, Alibaba's Claude Code ban, Crusoe's $3B raise talks and Fable 5 benchmark claims led 03 Jul.

OpenAI's proposed US stake, NVIDIA compute financing, DeepSeek API pricing, Z.ai's ZCode launch and Together AI's $800M round led 02 Jul.

Claude Fable access returns, Sonnet 5 pricing gets benchmarked, Etched exits stealth, and memory efficiency runs through today's AI stories.