A compact selection of competition mathematics problems supports repeatable reasoning checks.
83 model results
One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
96 % · MATH 500 · Accuracy · Cost In / Out: $1.25 / $10; Latency: 32.22s [2]
Published observations span 3.4 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#16 of 83 in this displayed comparison; 1.5 pp below the leader.
96.2 % · MATH 500 · Accuracy · Cost In / Out: $3 / $15; Latency: 27.26s [2]
Published observations span 2.8 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#18 of 83 in this displayed comparison; 1.6 pp below the leader.
94.6 % · MATH 500 · Accuracy · Cost In / Out: $2 / $8; Latency: 16.59s [2]
Published observations span 4.6 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#23 of 83 in this displayed comparison; 2.3 pp below the leader.
90.4 % · MATH 500 · Accuracy · Cost In / Out: $15 / $60; Latency: 23.54s [2]
Published observations span 6.6 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#40 of 83 in this displayed comparison; 5.5 pp below the leader.
Publisher-selected recent-model cohort from the public evaluation page. Model labels preserve effort and fallback configurations. Values are a dated snapshot, not a live feed. Consult the publisher’s methodology for the harness, task split and judging; do not merge this cohort with vendor-run results.
Public source leaderboard captured 26 October 2026. Model names, configurations, native score units and source-reported uncertainty are retained. Version and task split follow this specific Vals methodology page. These are published observations, not evaluations run by RLTasks.
TRACE THE SCORE · Per-model notes use published breakdowns where available.
What it measures
A compact selection of competition mathematics problems supports repeatable reasoning checks.
How to read the result
Pair final-answer scoring with a proof-sensitive evaluation if mathematical validity matters to the application.
Limits and comparison traps
Saturation and public-solution exposure limit how much a small gain says about research-level mathematics.
Published evaluation
The linked publication documents the task and scoring procedure. Model results retain the source’s evaluation setting, native score units and available uncertainty.
Where this fits in an evaluation suite
Final-answer accuracy, proof credit, and expert-reviewed research progress are distinct signals. A proof-aware evaluation should retain partial work and audit strong claims.
Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.
A reproducible comparison
Use the exact implementation, version, task subset, tool access, attempt budget, and judge described by the source.
Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.