Terra is priced at $1/M input and $5/M output tokens, a tenth of GPT-6 Astra's cost, with a 1.05M-token context and 128K max output. It scores 85.5% on MMMU-Pro vision at max effort, 1.2 points behind Astra, but only 50.3% on Terminal-Bench 4.0.
DeepSeek's update scores 45 on the Artificial Analysis Intelligence Index, up 6 points and tied for #2 among open-weights models. It jumps Terminal-Bench 4.0 from 31.2 to 42.3 and hits 77.0 on DeepSWE, above Opus 5 (74.0) and GPT-5.6 Sol.
The release covers 90 of the top 500 open problems per Proof of Atlas, organized into 372 families, with Lean formalizations on GitHub. Zvi Mowshowitz reports a single model averaged three hours of Pro-level compute per solution after being asked to try ~4,000 problems.
Anthropic filed a Form S-1 for a proposed IPO of Class A common stock, planning to list on the Nasdaq Global Select Market under the ticker "ANT"; share count and price range are not yet set. Latent Space reports a 1400% growth rate, 48% of revenue through cloud partners, 85% margin, $1B operating profit in Q2 and $134B cash.
Claude Duet is a research preview open by waitlist to Pro and Max subscribers in the US, powered by Opus 5.5. It runs on a cloud computer, messages you first with a morning brief, and waits for approval before sending, paying, or signing anything.
The open-weight MoE model has ~4B active parameters and handles 40 languages, speaker attribution for up to 8 speakers, and translation to and from 31 languages in one checkpoint. It tops Hugging Face's Open ASR Leaderboard for open models at 4.71% mean WER on 9 English sets, but works only on recordings, not live audio.