An agent completes technical work in a terminal under executable task checks.
124 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.
Published observations span 18.65 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#2 of 124 in this displayed comparison; 0.59 pp below the leader.
Published observations span 18.8 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#3 of 124 in this displayed comparison; 5.57 pp below the leader.
Published observations span 16.4 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#4 of 124 in this displayed comparison; 6.62 pp below the leader.
Published observations span 13.71 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#5 of 124 in this displayed comparison; 6.97 pp below the leader.
Published observations span 21.82 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#8 of 124 in this displayed comparison; 10.25 pp below the leader.
Published observations span 13.47 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#17 of 124 in this displayed comparison; 18.38 pp below the leader.
Published observations span 10.94 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#18 of 124 in this displayed comparison; 19.6 pp below the leader.
Published observations span 0 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#28 of 124 in this displayed comparison; 22.92 pp below the leader.
Published observations span 11.46 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#31 of 124 in this displayed comparison; 24.16 pp below the leader.
Published observations span 16.85 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#36 of 124 in this displayed comparison; 26.63 pp below the leader.
Published observations span 0.03 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#37 of 124 in this displayed comparison; 26.81 pp below the leader.
Published observations span 18.47 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#44 of 124 in this displayed comparison; 30.95 pp below the leader.
Published observations span 20.95 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#45 of 124 in this displayed comparison; 31.67 pp below the leader.
Published observations span 7.6 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#46 of 124 in this displayed comparison; 31.96 pp below the leader.
Published observations span 0.02 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#47 of 124 in this displayed comparison; 32.63 pp below the leader.
Earlier-2026 baseline cohort from the official model card. Asterisks identify competitor scores re-evaluated by Kimi rather than cited from their reports. Coding scores average ten runs; vision averages three runs. Tool-enabled and no-tools results, F1 and exact accuracy, pass@3 and pass^3 remain separate rows. MCPMark here predates the Verified default. APEX uses 452 public tasks.
Earlier-2026 baseline comparison from the official Qwen model card. Original model labels and score formatting are retained. Language and vision tables use different Qwen reference models. Tools, image inputs, agent scaffolds and task subsets can differ; follow the model card’s evaluation details. Video-MME subtitle settings and HLE tool settings stay in separate rows. Original MCP-Mark results predate MCPMark Verified.
Epoch AI public data snapshot, CC BY 4.0. Externally collected result; original evaluation provider is linked. CSV: terminalbench_external.csv; score column: Accuracy mean; fraction scale: 1. Model version is retained in configuration. Original source authors retain credit and any additional licensing terms.
Epoch AI public data snapshot, CC BY 4.0. Externally collected result; original evaluation provider is linked. CSV: terminalbench_external.csv; score column: Accuracy mean; fraction scale: 1. Model version is retained in configuration. Original source authors retain credit and any additional licensing terms.
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.
Score vs. release date
Epoch AI · 202 observations
Terminal Bench · Accuracy mean · % · higher is better
Dates are model release dates recorded by Epoch AI, not evaluation dates. Each point retains its published score and configuration. Separate releases are not a matched longitudinal experiment. Missing dates are omitted. Explore all reported results ↓
The dashed line traces the best observed score through the model release dates. It describes this selection of reports, with their original configurations. It does not estimate unmeasured models or costs.
View plotted values and sources (202)
Terminal-Bench 2.0 · Terminal Bench · Accuracy mean · %. Published measurements in their original score units.
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 (63)
Terminal-Bench 2.0 · Terminal-Bench 2.0 · 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.
An agent completes technical work in a terminal under executable task checks.
How to read the result
The model and agent harness form the evaluated system. Compare task completion with resource budgets, environment handling and recovery from failures.
Limits and comparison traps
Keep release 2.0 separate from 2.1, Hard and 4.0. Refreshed task verifiers and different scaffolds can change the measured outcome.
Epoch data and evaluation setting
Epoch AI publishes this benchmark in its capability collection. Imported rows retain the published score column and model configuration. Original authors and evaluation providers remain credited in the result references.
Epoch attribution and score coverage
Data is attributed to Epoch AI under CC BY 4.0, with original evaluation providers credited in the linked source snapshots. Scores are imported only where a published mapping identifies the field and its scale; missing or incompatible records are omitted rather than assigned zero.
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
The agent wrapper is part of the system. Pin tools, dependencies, timeouts, tests, and the repository revision. Compare contest coding separately from repository maintenance.
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
Pin the benchmark release, task split, model configuration, tools, sampling budget, and evaluation harness described by the primary 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.