Research workflows are completed in a terminal and checked with task-specific tests.
79 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.
63.33% · Terminal-Bench-Science 0.1 · AA harness · Publisher setting: Max [1]
Published observations span 4.77 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
Leads this displayed comparison.
Published breakdowns
Reported domain success ranges from 54.9% in physical sciences to 78.4% in mathematical sciences.
Published observations span 4.6 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#3 of 73 in this displayed comparison; 4.81 pp below the leader.
Published breakdowns
Reported domain success ranges from 49.1% in life sciences to 66.7% in mathematical sciences.
Reported domain success ranges from 42.1% in life sciences to 72.5% in mathematical sciences. #4 of 73 in this displayed comparison; 7.06 pp below the leader.
Reported domain success ranges from 40.7% in engineering sciences to 72.5% in mathematical sciences. #5 of 73 in this displayed comparison; 7.54 pp below the leader.
52.86% · Terminal-Bench-Science 0.1 · AA harness · Publisher setting: High [1]
Published observations span 4.74 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#6 of 73 in this displayed comparison; 9.93 pp below the leader.
Published breakdowns
Reported domain success ranges from 41.7% in earth sciences to 70.6% in mathematical sciences.
Reported domain success ranges from 29.2% in earth sciences to 64.7% in mathematical sciences. #7 of 73 in this displayed comparison; 11.83 pp below the leader.
Published observations span 9.27 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#9 of 73 in this displayed comparison; 14.87 pp below the leader.
Published breakdowns
Reported domain success ranges from 31.6% in life sciences to 51.0% in physical sciences.
Reported domain success ranges from 29.6% in engineering sciences to 39.2% in mathematical sciences. #14 of 73 in this displayed comparison; 28.02 pp below the leader.
Reported domain success ranges from 22.2% in engineering sciences to 43.9% in life sciences. #16 of 73 in this displayed comparison; 30.87 pp below the leader.
Published observations span 1 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#18 of 73 in this displayed comparison; 35.66 pp below the leader.
Reported domain success ranges from 18.5% in engineering sciences to 33.3% in earth sciences. #20 of 73 in this displayed comparison; 39.44 pp below the leader.
Reported domain success ranges from 12.3% in life sciences to 29.2% in earth sciences. #24 of 73 in this displayed comparison; 45.16 pp below the leader.
Reported domain success ranges from 7.8% in mathematical sciences to 25.9% in engineering sciences. #28 of 73 in this displayed comparison; 52.3 pp below the leader.
Reported domain success ranges from 5.3% in life sciences to 23.5% in physical sciences. #29 of 73 in this displayed comparison; 53.25 pp below the leader.
Reported domain success ranges from 3.7% in engineering sciences to 15.8% in life sciences. #31 of 73 in this displayed comparison; 55.16 pp below the leader.
Reported domain success ranges from 3.9% in mathematical sciences to 19.6% in physical sciences. #36 of 73 in this displayed comparison; 55.63 pp below the leader.
Reported domain success ranges from 0.0% in earth sciences to 13.7% in physical sciences. #37 of 73 in this displayed comparison; 56.59 pp below the leader.
Reported domain success ranges from 0.0% in mathematical sciences to 16.7% in earth sciences. #40 of 73 in this displayed comparison; 59.44 pp below the leader.
Reported domain success ranges from 0.0% in physical sciences to 16.7% in earth sciences. #45 of 73 in this displayed comparison; 60.4 pp below the leader.
Reported domain success ranges from 2.0% in mathematical sciences to 11.1% in engineering sciences. #45 of 73 in this displayed comparison; 60.4 pp below the leader.
Terminal-Bench-Science 0.1 · AA harness · 2026-10-28
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.
Reported values are percentages except GDPval-AA, which is Elo. Tool-enabled and partial scores keep their labels. Effort settings, fallback models, and source-run protocols vary; see the original footnotes.
A transcription of the October model card, including its external leaderboard snapshots. These are not uniformly matched runs. Effort, tools, hardware, frame budgets and task subsets differ. GraphWalks reports F1; OSWorld uses the offline subset and partial credit. Harvey refers to the archived Vals v1 protocol, not the remediated v1.1 or AA variant. Consult the linked methodology.
Launch-reported maxima across reasoning efforts, not a uniformly matched effort comparison. Research/API settings can differ from ChatGPT. Source footnotes disclose altered harnesses, model revisions and reproduced competitor results. A dagger marks cells with additional source qualifications. Preserve each row’s named subset and tools.
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.
Terminal-Bench-Science 0.1 · Kimi K3.1 Pro · complete release comparison · 2026-10-27
Published model comparison, 2026. Kimi K3.1 Pro · complete release comparison. Exact values from the publisher’s embedded full comparison table. Maximum effort, fallback settings and dated model versions remain in model labels. Blank cells stay missing. Source footnotes specify mixed harnesses and externally quoted comparisons; this is a launch-reported cohort rather than a new independent evaluation. Benchmarks · full embedded comparison table · row 2; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
TRACE THE SCORE · Per-model notes use published breakdowns where available.
Score vs. cost per test
Vals AI · 41 observations
Terminal-Bench-Science · Score · higher is better
Costs are Vals AI’s published Cost / Test in USD, paired with the score from the same result. They reflect that evaluation’s workload and configuration; they are not token prices or cost per successful task. Missing costs are omitted. Explore all reported results ↓
The dashed line traces the best observed score as reported cost increases. It describes this selection of reports, with their original configurations. It does not estimate unmeasured models or costs.
View plotted values and sources (41)
Terminal-Bench-Science 0.1 · Terminal-Bench-Science · Score. Published measurements in their original score units.
These model reports include this benchmark. Their harnesses, task subsets, scoring policies, and resource budgets can differ; follow each report’s evaluation notes before joining the numbers.
Research workflows are completed in a terminal and checked with task-specific tests.
How to read the result
This complements question answering by asking whether a model can construct and execute a scientific workflow.
Limits and comparison traps
Executable checks establish the stated task conditions; passing them alone does not validate every scientific interpretation in a submission.
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.
Published evaluation
Exact values from the publisher’s embedded full comparison table. Maximum effort, fallback settings and dated model versions remain in model labels. Blank cells stay missing. Source footnotes specify mixed harnesses and externally quoted comparisons; this is a launch-reported cohort rather than a new independent evaluation.
Where this fits in an evaluation suite
A correct multiple-choice answer and a supported scientific contribution demand different evidence. Check whether the score rewards a proposal, an executed experiment, or a completed research target.
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
Exact values from the publisher’s embedded full comparison table. Maximum effort, fallback settings and dated model versions remain in model labels. Blank cells stay missing. Source footnotes specify mixed harnesses and externally quoted comparisons; this is a launch-reported cohort rather than a new independent evaluation.
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.