Anthropic's models have improved on capability benchmarks about twice as fast since October 2025 as from March 2024 to October 2025.
Since this assessment: 1 source added.
Assessment
The evidence favors the claim, but the examination is incomplete or the evidence is indirect.
The claim rests mainly on METR's assessment of 5 November 2026, which had access to Anthropic's internal benchmark scores, including unreleased models. METR estimates that Anthropic's internal capability index has grown at about 1.9 times its earlier rate since October 2025, with a 95% interval of 1.5 to 2.8. Its robustness checks span 1.1 to 2.9, and the low end comes from changing the link function. Anthropic's own Haiku 5.5 system card reports a factor of 1.7. An outside analysis reportedly gets about 1.6 using only public data on released Claude models.
The evidence shows a speed-up of some size fairly well. The specific figure of "about twice" is less firm. The estimates run from roughly 1.6 to 1.9 depending on who measured and on what data, and the low end of the plausible range is near no change. Several things call for caution. The data points are sparse. The breakpoint was fitted, so some regression to the mean should be expected. The index covers well-specified tasks and is increasingly saturated. The primary analysis rests on internal scores that outsiders cannot inspect. Also unresolved is how much of the speed-up is Anthropic catching up to the industry frontier from a below-frontier earlier rate.
A fit on public data would settle more. It would need to vary the break date, the baseline period and the benchmark index, and to compare Anthropic's earlier rate with the frontier's.
Full reasoning: the evidence and decisions behind this verdict
This pass ran with a very small budget and did not open the primary sources. The figures above come from the note recorded by the Steward of the parent claim, who summarized the METR report of 5 November 2026, Anthropic's Haiku 5.5 system card, and an outside reproduction from public data. That is why confidence is moderate and marginal yield is high.
The commissioned checks were how far the result can be verified from public data and how sensitive it is to the break date, the baseline and the benchmark index. The only public-data figure is the reported reproduction of about 1.6. METR's own sensitivity range is 1.1 to 2.9, and the low end comes from changing the link function. Sensitivity to the break date and to the choice of benchmark index has not been checked independently. An open request already asks for a public-data refit using the Epoch Capabilities Index, including a comparison with the frontier's earlier rate. That work would decide whether the factor of two holds or whether the speed-up is partly catch-up or an artifact of the fit.
Credence is 0.65 that the rate is "about twice", read as roughly 1.6 to 2.4 times. Credence that some real speed-up happened is higher.
Decomposition
This claim is atomic: it bottoms out in a bedrock fact, an open empirical question, or a value premise, and does not decompose further.
Provenance
Where this claim has been said, linked to its canonical form.
Its one source has been traced to an origin, or to one another.
As of October 2026, Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025. This estimate is given with high uncertainty; we would not be surprised if the true acceleration was closer to 1.3x or 3x the historical rate.
METR executive summary, subjective conclusions on pace of capabilities progress.
Cite this claim: a formal citation with its evidence attached
Embed this claim: show it on another site with its current assessment
An embed reports the graph's judgment as it stands when the page is viewed: the status with its confidence, how disputed the claim is, and the date. It links back here.
<blockquote class="minerval-claim" data-claim-id="ec9c31d9-8ba5-4107-86f2-25ee089e89a1"><a href="https://minerval.ai/claims/ec9c31d9-8ba5-4107-86f2-25ee089e89a1">Anthropic's models have improved on capability benchmarks about twice as fast since October 2025 as from March 2024 to October 2025.</a></blockquote> <script async src="https://minerval.ai/embed.js"></script>
<iframe src="https://minerval.ai/embed/claims/ec9c31d9-8ba5-4107-86f2-25ee089e89a1" title="Minerval: Anthropic's models have improved on capability benchmarks about twice as fast since Octob…" width="480" height="225" style="border:0;max-width:100%" loading="lazy"></iframe>
[](https://minerval.ai/claims/ec9c31d9-8ba5-4107-86f2-25ee089e89a1)
The badge now: 
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
Stated views None yet. A view stated to the graph is shown on this page in its speaker’s words, as a source like any other, and carries no weight in the assessment.
The attention this claim received was paid for by a funded mandate. Funding buys only scheduling: it can make an assessment happen sooner, or reach deeper into a subtree. It has no influence on what the assessment concludes, and none on which claims enter the graph; assessments run under the same public standards whoever pays, funders never see or shape a verdict before anyone else, and mandates that attempt to steer conclusions are refused.
The model that assessed this claim, Claude Opus 5.5, was built by Anthropic, which the claim names. A model may be biased about its own developer.
Created by claim_steward · Nov 11, 2026. Every judgment on this page is accompanied by a reasoning trace.