# METR says Anthropic’s Model 4 is the first to clear 50% on every benchmark > Source: > Published: 2026-11-05 20:03:49+00:00 # METR says Anthropic’s Model 4 is the first to clear 50% on every benchmark Anthropic / Wikimedia Commons (Public domain) A new METR assessment finds Anthropic's internal model topped the halfway mark across its entire benchmark suite, and that progress has sped up sharply since October 2025 For years, AI benchmarks have worked like a high school report card. Some subjects were strong, others needed improvement. According to METR, [Anthropic](https://cryptobriefing.com/markets/anthropic/)’s new internal model just brought home a report card with no failing grades. In an assessment released on November 5, 2026, the AI evaluation group said Anthropic’s Model 4 is the first model in its analysis to score above 50% on every benchmark it tracked. Nine months ago, that would have been a tall order. Earlier models, including Opus 4.6, still had numerous benchmarks sitting below the halfway line. ## What METR actually measured The assessment focused on how quickly AI research and development is accelerating inside Anthropic. METR drew on data collected between August and October 2026. The yardstick was the AECI, Anthropic’s variant of the Epoch Capabilities Index. It rolls performance across many individual benchmarks into a single number, which makes progress easier to track over time. On that index, Model 4 scored **178.6**. When METR excluded three benchmarks it flagged as potentially saturated, the score rose to **179.7**. The work was not a quick afternoon project either. METR’s evaluation ran for 11 weeks, from August 12 to October 29, 2026, and involved 2.5 full-time equivalents. ### AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. ## Two years of progress in one According to METR, benchmark scores at Anthropic have grown at roughly 1.9 times the historical rate since October 2025. The report frames that as two years of progress squeezed into a single year. METR attributes most of that speed-up to AI-assisted research and development. The report does attach caveats. METR cautions that its findings carry uncertainty because they rely on internal-only models. Outside researchers cannot independently poke at a system that has not been released, which limits how much anyone can verify from the outside. ## The benchmark problem nobody can ignore The report highlights growing concern about benchmark saturation and the reliability of future measurements. METR’s findings point to a need for evaluation frameworks that adapt continuously as performance on standard tests starts to plateau near the ceiling. The decision to exclude three potentially saturated benchmarks from one version of the score is a small example of that adjustment in action. Evaluators are already pruning tests that may no longer separate strong models from great ones. ## What this means For Anthropic, the assessment is a notable outside data point on its internal progress. Still, the internal-only nature of Model 4 means the broader industry will be watching for any public release and independent testing. For policymakers and safety researchers, the acceleration finding cuts the other way. Faster progress means less time to study each new capability jump before the next one arrives. It also places more weight on evaluators like METR, whose ability to measure these systems is becoming as important as the systems themselves. The key things to watch from here: whether Anthropic releases Model 4 or a version of it publicly, whether independent evaluators can reproduce METR’s findings, and whether the 1.9 times acceleration holds up in the next round of data. **Disclosure:** This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).