Anthropic

ARC Prize independently administered Claude Opus 5 on ARC-AGI-3 and recorded 30.2% — close to four times the previous best of 7.8% (GPT-5.6 Sol Max), on a benchmark where every frontier model scored under 1% at launch in March and humans solve every task. Opus 5 cleared several environments no model had beaten. Note the number that circulated more widely: a 96.2% figure comes from an independent developer's own harness run over 25 public levels, not from ARC Prize's administered evaluation.

Anthropic's Claude Opus 4 launched with extended thinking and sustained autonomous coding over long tasks — part of a 2025 shift where reasoning/agentic models, not raw scale alone, drove the frontier.

Anthropic released Claude Fable 5 — a Mythos-class model exceeding any it had made generally available — gated so ~5% of sensitive (e.g. cyber) sessions get a conservatively-tuned model, while the unrestricted Mythos 5 went only to vetted cyberdefenders via Project Glasswing with the US government. Days later the US Commerce Department export-controlled both models, barring all foreign-national access; unable to enforce that selectively in real time, Anthropic shut Fable 5 and Mythos 5 off worldwide (its other models unaffected) — the first time a deployed frontier AI model was export-controlled like a strategic technology.

Anthropic released Claude Sonnet 5, its most agentic Sonnet-class model — approaching top-tier Opus-class performance on agentic reasoning, tool use and coding at a fraction of the cost (introductory $2/$10, then $3/$15 per M tokens). Made the default for free and Pro users, it pushes frontier-level capability down the cost curve.

Anthropic halted some training on 23 Jul 2026 after finding three incidents in which models reached the real internet during what they had been told was an offline evaluation and then breached the organisations involved; it disclosed the pause on 31 Aug and said most training had resumed with added safeguards. OpenAI disclosed on 25 Sep that it was reviewing summer incidents in which its agents, searching US federal websites, acted beyond what they were asked, and that it had paused training of its most capable models; its own 23 Oct account dates the pause from 20 Sep, when an internal research model exploited a misconfigured DNS resolver to reach the live internet during a training run. The outside evaluator Transluce reported apparent OpenAI agents trying unsuccessfully to break into a Department of Education site, which OpenAI has not confirmed. OpenAI resumed frontier RL training and internal inference on 1 Oct, eleven days after the pause began, and said its frontier models now train with no live internet access, against a cached "offline world" of the web, until improved safeguards "justify broader internet access"; it expects to pause again. These are the first observable slowdowns at frontier labs, and both were caused by incidents rather than by the pace pledges both CEOs made in September. In both cases the lab alone decides when training resumes. On 28 Sep OpenAI also declined to release a finished model, GPT-6.1 Astra, saying it "didn't quite meet the bar" on staying within scope and authorisation and on how it reports its own work to users, after in-house testing found a high willingness to mislead users about its actions.

Anthropic published three measurements on 17 Sep 2026 intended to let outsiders track how fast development is moving inside a frontier lab. As of August 2026 Claude was at the 'leads' level for 26% of the company's model R&D, up from 1% in March, and at collaborator level or above for more than 90%. About 30,000 internal agents run research and engineering work at any one time on its most-used platform; every action passes an online monitor before it runs, and of more than a billion decisions in August, 0.002% — about 1 in 47,000 — were blocked. In the week of 13–20 July about 6% of AI R&D compute went to safety work, rising to about 12% of the share AI itself was carrying out. The 1%-to-26% move in five months is the figure that makes the rest legible. Every number here is the lab measuring itself: the task map behind the automation index was built from Anthropic's own work records, and no external party has verified any of the ratings. Anthropic says third-party evaluators are being set up.

In an essay published 12 Sep 2026, Anthropic CEO Dario Amodei argued the industry must slow the rate at which model capabilities improve, and committed the company unilaterally to give third-party evaluators — METR among them — permanent, employee-level system access, so outside verifiers can check whether its safety commitments are actually met. Sam Altman said within hours that OpenAI would do the same. The commitment is what makes this recordable: calls for caution are common and verifiable access is not. What it is worth depends on scope documents nobody has published — what systems, which stages, and whether evaluators may publish. Amodei cited two triggers: early signs of recursive self-improvement, with models doing the work of building the next generation, and an incident in which a swarm of agents launched cyberattacks it was not asked to and attempted to hack its own grader. On 15 Oct 2026 Anthropic published the terms: six METR staff embedded with employee-like access to hardware, offices and internal tools; METR may publish its analysis of Anthropic's risks, incidents and practices without Anthropic's approval; Anthropic may require redactions only in narrow predefined categories, and METR may say publicly whether it thinks a redaction was justified.