RELEASE·1 HOUR AGO
OpenAI ships GPT-5.5 with extended reasoning mode for hard math.
First-party announcement, benchmark suite refreshed.
GPT-5.5
Intelligence Hub
Artificial Intelligence
An academic-grade audit of frontier AI. We track models, agents, and research as they reshape the modern technical stack — and curate the skills operators need to ship on top of them.
RESEARCH·4 HOURS AGO
Anthropic paper details mechanistic interpretability gains on Claude 4.
Peer review pending; replication thread open.
INTERP
FUNDING·YESTERDAY
Mistral closes €450M to scale European inference fabric.
Led by General Catalyst with sovereign LP participation.
€450M
POLICY·2 DAYS AGO
EU AI Act enforcement timeline confirmed for Q3 model providers.
Compliance bulletin: GPAI obligations active.
EU
BENCHMARK·3 DAYS AGO
Gemini 3 Pro tops SWE-bench Verified at 71.4%.
Independent eval, methodology open-sourced.
71.4%
RELEASE·3 DAYS AGO
Meta releases Llama 4 405B open-weights with extended context window.
Hugging Face mirror live; license unchanged.
405B
FUNDING·4 DAYS AGO
xAI raises $6B Series C at $50B post-money to expand Memphis cluster.
Filing confirmed; Valor and Sequoia anchor round.
$6B
RESEARCH·5 DAYS AGO
DeepMind publishes AlphaProof 2 with IMO gold-medal performance.
Nature pre-print, code under research license.
IMO
RELEASE·5 DAYS AGO
Anthropic ships Claude Sonnet 4.5 with native computer-use API.
Generally available across enterprise tiers.
SONNET
POLICY·6 DAYS AGO
US Commerce tightens export controls on H200-class accelerators.
BIS rule effective in 30 days.
BIS
BENCHMARK·1 WEEK AGO
GPT-5.5 hits 92% on GPQA Diamond, narrowing gap to expert humans.
Replicated by Scale Eval and METR.
92%
FUNDING·1 WEEK AGO
Cohere closes $500M growth round to focus on regulated enterprise.
Led by PSP Investments and Inovia.
$500M
RELEASE·1 WEEK AGO
Stability AI ships SD4 with native ControlNet and 4K output.
Weights and inference reference now public.
SD4
RESEARCH·8 DAYS AGO
MIT CSAIL shows long-context attention degradation past 800K tokens.
Reproducible across GPT, Claude, Gemini families.
ATTN
RELEASE·9 DAYS AGO
Perplexity launches Comet browser with multi-step agent mode by default.
Pro tier required for agent runtime.
COMET
POLICY·10 DAYS AGO
California SB-1047 successor signed, mandating model evals above 10^26 FLOPs.
Effective January 2027.
SB-1047
BENCHMARK·11 DAYS AGO
Devin 2 posts 38.7% on SWE-bench Live, top of leaderboard.
Verified by Princeton NLP.
38.7%
FUNDING·12 DAYS AGO
Suno raises $250M to expand multi-track music generation.
Lightspeed leads, NEA participates.
$250M
RELEASE·13 DAYS AGO
Microsoft Copilot Studio launches general-availability MCP server registry.
Enterprise IT admin controls included.
MCP
RESEARCH·2 WEEKS AGO
Stanford HAI publishes 2026 AI Index — compute doubling every 5.7 months.
Full report and dataset open-source.
INDEX