- Deep Think reasoning rolls out to 2.5 Pro on the Gemini API, AI Studio, and Vertex AI — no longer Ultra-only.
- Project Mariner's browser-control tool and native audio output now exposed for both 2.5 Pro and 2.5 Flash.
- Thinking budgets extend to 2.5 Pro so developers can cap or disable reasoning tokens per call; structured thought summaries land in the response payload.
- Cited benchmarks: 87.6% LiveCodeBench, 65th percentile USAMO, IMO gold-medal standard.
- Full GPT-5.5-Cyber GA for verified defenders: 85.6% on CyberGym vs 81.8% for stock GPT-5.5.
- Codex Security plugin ships in the Codex app — finds, validates, and patches vulnerabilities inline with attack-path traces and severity reports.
- 'Patch the Planet' partners with Trail of Bits, HackerOne, and Calif across cURL, Python, Go, urllib3, PyPI, Valkey, RustCrypto, and 12 more.
- First-week tally: hundreds of bugs, 64 PRs, 51 issues across 19 projects; field results include a Firefox WASM CVE patched before Pwn2Own.
- Nvidia-backed Reflection AI commits $150M/month from July 1 through 2029 for GB300 capacity at SpaceX's Memphis data center.
- Reflection has yet to ship a public frontier model; this is the compute it needs to actually train one.
- Adds an open-source lab to Colossus's roster alongside existing Anthropic, Google, and Cursor deals.
- SPCX closed off ~10%, its worst day since the Nasdaq debut — the market still hasn't decided if AI compute revenue makes SpaceX a hyperscaler.
- Nous Research's self-evolving agent runtime added 23K stars in roughly two weeks; Dealroom names it the fastest-growing open-source agent framework of 2026.
- June 5's 'Surface Release' v0.16.0 shipped a native macOS/Windows/Linux desktop app with one-click install and concurrent multi-profile sessions.
- Skills Hub now lists 90,881 community skills across 12 registries, up from a few hundred at launch.
- Self-evolution loop pairs DSPy + GEPA to let Hermes optimize its own prompts, skills, and code from real usage.
- AlphaFold lead and 2024 Chemistry Nobel winner is leaving Google DeepMind after almost nine years to join Anthropic after a short break.
- Lands four days after Gemini co-lead Noam Shazeer's jump to OpenAI — two top architects out of DeepMind in one week.
- Anthropic gets a marquee science-AI hire on top of its existing Claude-for-Science push and the John Jumper-style protein work it already funds.
- Bloomberg broke the story; CNBC, TechCrunch, and TNW confirmed within hours.
01
Gemini 2.5 Pro Gets Deep Think, Computer Use, and Native Audio in One Drop
models deepmind.google
02
OpenAI Opens GPT-5.5-Cyber to All Defenders and Sics Codex on Open-Source CVEs
tools openai.com
03
SpaceX Sells Reflection $6.3B of Colossus 2 Compute for an Open-Weights Frontier Run
industry cnbc.com
04
Hermes Agent Crosses 200K Stars With Skills Hub Past 90K
open-source github.com
05
Nobel Laureate John Jumper Defects From DeepMind to Anthropic
industry bloomberg.com