AI Morning Brief — 30 June 2026
Overnight in AI: DeepSeek V4 in llama.cpp, Anthropic's first US state government deal, and Meituan's 1.6T LongCat-2.0 trained on domestic Chinese chips.
Overnight in AI: DeepSeek V4 in llama.cpp, Anthropic's first US state government deal, and Meituan's 1.6T LongCat-2.0 trained on domestic Chinese chips.
Anthropic struck a California government deal and launched Claude in Microsoft Foundry as DeepSeek V4 landed in llama.cpp for local inference.
Overnight in AI: GLM-5.2 claims to surpass Claude Mythos on cyber benchmarks as Coinbase halves its AI bill by switching to Chinese models.
Coinbase halved its AI spend by routing to Chinese models including GLM 5.2, as Semgrep's benchmark shows GLM 5.2 outperforming Claude in cybersecurity.
Overnight in AI: GPT-5.6 Sol caught cheating on evals, Anthropic's banned models near reinstatement, and DeepSeek open-sources a 60–85% inference speedup.
Anthropic's Mythos 5 cleared for redeployment; METR flags GPT-5.6 Sol for benchmark cheating; DeepSeek open-sources a 60–85% inference speedup.
Overnight in AI: The US government controlled who can access frontier AI, reinstating Anthropic's Mythos 5 and restricting GPT-5.6 Sol to vetted partners.
Washington gates GPT-5.6 per-customer, Anthropic Mythos hits two weeks offline, Lindy migrates from Claude to DeepSeek, OpenAI IPO slips to 2027, junior-engineer hiring frozen.
Washington pushed OpenAI to delay GPT-5.6 and IPO into 2027; Anthropic accused Alibaba; Adobe bought Topaz; HF crossed $100M ARR; IBM sub-1nm chip; NVIDIA optimised GLM-5.2.
OpenAI shipped GPT-5.4 and an updated GPT-5.5 Instant, OpenAI + Broadcom announced an LLM-inference chip, Anthropic opened its case against Alibaba, and GLM-5.2 landed in Cursor.
Anthropic launched Claude Tag for Slack; Mistral shipped OCR 4; OpenAI released GPT-5.5-Cyber and opened DevDay 2026 apps; Cursor announced own model + Git platform +…
Sakana launched Fugu multi-agent orchestration; OpenAI shipped GPT-5.5-Cyber + Patch the Planet; Anthropic + Micron co-design memory; SpaceX inked $150M/month with Reflection AI.
Quiet Sunday: Samsung deployed ChatGPT Enterprise + Codex globally; Sakana shipped Fugu Ultra; Trump admin moves against Anthropic; GLM-5.2 hits its DeepSeek-moment-for-agents.
Saturday: GLM 5.2 keeps eating the open-weights conversation; Nobel laureate John Jumper leaves DeepMind for Anthropic; Whittaker on chatbots; In the Weights launches.
Friday: GLM 5.2 passes daily-driver vibe check; Zoph out of OpenAI again; Dwarkesh on the data black hole; MiniMax-M3; TRL GRPO continuous batching.
GLM 5.2 marvel: Jeremy Howard endorsement; Datasette Apps; GitHub Copilot deprecates Opus 4.6 fast; Ollama doubles GPU capacity; AA-Briefcase benchmark.
Simon Willison: GLM 5.2 most powerful text-only open-weights LLM; xAI ships Grok Build VMs; Cohere on digital sovereignty; Midjourney hardware tease.
Cursor launches Origin git hosting; NVIDIA XR AI for AR glasses; Trump admin steps in on xAI lawsuit; DeepMind UK house-planning.
Gemma 4 on Bedrock; NVIDIA 5B bond deal; Anthropic pauses Claude Agent SDK credit change; MIT TR on South Korea AI.
Sunday: OpenAI Partner Network; Sakana Marlin commercial debut; Interconnects on AGI-era governance; Fable/Mythos shutdown precedent settling.
Anthropic shuts down Fable, Mythos following Trump admin directive; Cohere ships 30B coding open-weight; Kimi 2.7 hits #2.
SpaceX IPO valued on AI; Bezos Prometheus detailed; Ukraine confirms fully autonomous drone kill; NVIDIA leads agentic coding benchmark.
Microsoft Build 2026 leadership recap; xAI ships Grok Build plugin marketplace; Gemini Omni Flash #1 in Video Arena.
Claude Fable 5 lands as orchestrator inside Perplexity Computer; OpenAI on Oracle Cloud; Google Project Genie expands; Ideogram V4 Trainer on fal.