Share of Anthropic internal Claude Code sessions on the most open-ended tasks judged successful in May 2026 (%)
- Measures
- Percentage of Claude Code sessions by Anthropic staff, in Anthropic's highest task-difficulty tier ("most open-ended tasks", problems with no clear specification), that an automated Claude judge classed as successful, meaning the agent clearly succeeded at the user's tasks without requiring corrections.
- Population
- Internal Anthropic Claude Code sessions assigned by Anthropic to its most open-ended difficulty tier
- Conditions
- Calendar month of May 2026; success as judged by Anthropic's Claude-based session judge, with tiering and judging as Anthropic implemented them internally
- Unit
- a share, shown as a percentage
Between 7,100 and 8,000% (90% interval), most likely about 7,600%.
Stated as an interval, the precision the evidence supports. centred on the reported 76%; width reflects rounding, how the May point was read from a noisy monthly series, and possible later revision of the internal pipeline
This estimate rests in part on evidence not shown: at least one of its sources reports a result from data it has not published.
Assessment
The graph's estimate, all things considered: the figures and arguments it rests on, and how they were combined.
About 76%, with a 90% interval of roughly 71% to 80%. The only source is Anthropic's own report "When AI builds itself", which states that on the most open-ended tasks Claude's success rate reached 76% in May 2026, up 50 percentage points in six months. Success is graded by a Claude-based judge: a session counts if the agent clearly succeeded at the user's tasks without requiring corrections. Anthropic is the only party that holds the session logs and runs the judge, so its figure is direct testimony about the value as defined here, and a specific dated number in a published report would be costly to misstate. The interval is kept somewhat wide because the underlying data, the rule assigning sessions to the most open-ended tier, and any validation of the judge are unpublished, and the report itself notes that shifting workloads cause short-term fluctuations, so a single month's point depends on how the series was smoothed or read. The figure measures what the internal judge recorded. It does not establish how often such sessions would be judged successful by human reviewers, or how the tier compares with external benchmarks; those questions bear on what the number means, not on its value.
The work: which figures, how they were combined, and the arithmetic
Single source, read in full: the Anthropic Institute report (Favaro and Clark, with a September 2026 update). Passage: "On the most open-ended tasks, Claude's success rate reached 76% in May 2026, up 50 percentage points in six months." Method note in the report: Claude judge; success means clear success without corrections; workload changes cause short-term fluctuations. No independent figure exists; no examination of the data by another party is reported. Central value taken as the reported 76. Interval: allow about +/-1 for rounding, a few points for which month-level or smoothed value the text reports versus the plotted series, and a few points for pipeline or tier-definition revisions (the page was updated after publication). Combined roughly as a 90% interval of 71 to 80, slightly wider below because self-reported highlight figures tend to be chosen from favourable readings of a noisy series.
Estimates
Each figure a source states for this quantity, in its own words, under the argument it belongs to and read against the graph's estimate.
Grey band: the graph's estimate, with its central value as a line. Dark marks: each source's own figure. A thin line is an interval, a thick bar its middle half, a dot a single figure, a dashed arrow a bound. An open box is where an argument places the value.
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 quantity's evidence comes from. The figures sources state are under Estimates above.
Its one source has been traced to an origin, or to one another.
On the most open-ended tasks, Claude’s success rate reached 76% in May 2026, up 50 percentage points in six months.
Success judged by a Claude judge on internal Claude Code sessions; a September 2026 update chart shows open-ended success near 91%.
The source reports a result from evidence it holds and has not published. The figure comes from Anthropic's internal Claude Code session logs, graded by a Claude-based judge that counts a session as successful when the agent clearly succeeded without requiring corrections. Neither the session data, the rule assigning sessions to the most open-ended tier, nor any validation of the judge against human grading is published, and no outside party is reported to have examined them. The report itself warns that shifting workloads cause short-term fluctuations in the rate. A later update to the page (September 2026) extended the chart beyond May.
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="3f34c80f-2078-4455-9b5f-fc4017565ce9"><a href="https://minerval.ai/claims/3f34c80f-2078-4455-9b5f-fc4017565ce9">Share of Anthropic internal Claude Code sessions on the most open-ended tasks judged successful in May 2026 (%)</a></blockquote> <script async src="https://minerval.ai/embed.js"></script>
<iframe src="https://minerval.ai/embed/claims/3f34c80f-2078-4455-9b5f-fc4017565ce9" title="Minerval: Share of Anthropic internal Claude Code sessions on the most open-ended tasks judged succ…" width="480" height="275" style="border:0;max-width:100%" loading="lazy"></iframe>
[](https://minerval.ai/claims/3f34c80f-2078-4455-9b5f-fc4017565ce9)
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.
Created by extractor · Nov 13, 2026. Every judgment on this page is accompanied by a reasoning trace.