Today in DeAI: Dario Amodei publishes a three-step plan to slow frontier AI and commits Anthropic to embedded third-party evaluators, Cognition ships a Kimi K3 post-trained coding model, and Cohere releases an open-weight translation model under a non-commercial license.
Amodei tells frontier AI labs to slow down, opens Anthropic to auditors
Anthropic CEO Dario Amodei published "We Must Pace the Frontier" on Saturday, a three-step proposal to slow the rate of frontier AI capability gain. The first step — permanent embedded third-party evaluators with employee-level access — is one Anthropic is committing to unilaterally, effective immediately. Sam Altman and Elon Musk both publicly endorsed the move within hours, per Reuters. Why it matters: the proposal creates a new verification axis between closed labs (audited by embedded evaluators) and open-weight models (audited by anyone), and regulators will have to pick a side. (essay) — our coverage
Cognition ships SWE-2, a Kimi K3 post-trained coding model
Cognition released SWE-2, post-trained from Moonshot AI's 2.8-trillion-parameter Kimi K3 with reinforcement learning, per MarkTechPost. Cognition reports 50.0% on FrontierCode 1.1 Main — within one point of Fable 5.1 at 64% lower cost — and leads on Terminal-Bench 2.1 at 92.8%. Why it matters: a coding lab post-training an open-weights base to within one point of a closed frontier model is the strongest evidence this month that open weights are commercially competitive at the application layer. SWE-2 itself is closed — it runs only inside Devin — but its base is not.
Cohere releases North Small Translate under CC BY-NC
Cohere released North Small Translate, a 218B-parameter mixture-of-experts translation model covering 50+ languages, with open weights under a CC BY-NC 4.0 license. Cohere reports an 83.6 score on WMT26 across all languages, ahead of DeepL and Google Translate in the same evaluation. Why it matters: it is the latest data point in the open-weights-versus-open-source licensing debate — the weights are downloadable, but commercial use requires a separate license, which limits who can build on it.
Nvidia in talks to invest up to $10 billion in Anthropic's IPO
Nvidia is in discussions to invest up to $10 billion in Anthropic's planned IPO at a $2 trillion valuation, per The Decoder. Why it matters: the same week Amodei proposes slowing frontier AI, his company is preparing what would be the largest IPO in history — and its biggest chip supplier is buying in. The tension between the safety pitch and the capital raise is the story to watch.
Watching tomorrow
Whether OpenAI or xAI convert their public endorsements of Amodei's evaluator plan into actual programs with the same publishing rights Anthropic has promised — and whether any open-weight lab responds with a formal counter-proposal.
Sources
- We Must Pace the Frontier — Dario Amodei
- Anthropic CEO urges AI companies to slow model development amid fears over misuse — Reuters
- Cognition Releases SWE-2: A Kimi K3 Post-Trained Coding Model — MarkTechPost
- Introducing North Small Translate — Cohere
- Nvidia in talks to invest up to $10 billion in Anthropic's IPO — The Decoder
About DeAI
DeAI is an independent publication covering open-weight AI models, private inference, and decentralized infrastructure — the tools for running AI you actually control. We test providers on price, privacy, and refusal behavior and publish the numbers, not the vibes. DeAI is powered by Morpheus (mor.org), a decentralized inference marketplace, and covers it on the same terms as every other provider.
Powered by Morpheus and StrandCMS
Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider — we rank it wherever the data lands. StrandCMS is the open-source, agent-first framework this site is built on.