TeamBench - Full Team: leaderboard

TeamBench (2026) tests agent coordination under operating-system-enforced role separation on a 90-task stratified subset graded by deterministic scripts, every role filled by the same model at temperature 0. In the full team a Planner reads the specification, an Executor edits the workspace and a Verifier certifies the result; pass rate in percent, higher is better.

Metric: Pass rate (%) on the 90-task stratified TeamBench leaderboard subset (operating-system-enforced role separation in Docker sandboxes; a run passes when every check of the task's deterministic shell-script grader passes, and a run failing only the attestation check counts as a pass), temperature 0, with a Planner (reads the specification), an Executor (edits the workspace) and a Verifier (writes the final attestation), all the same model; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.737.8#85
2Gemini 3.1 Pro (Preview)28.9#99
3Claude Haiku 4.528.9#583
4GPT-5.4 Mini28.9#406
5Claude Sonnet 4.627.8#173
6GPT-5.427.8#135
7Gemini 3 Flash25.6#175
8Gemma 4 31B22.2#382
9Gemini 3.1 Flash Lite17.8#371

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=teambench-full-team · How It Works · Data refreshed daily, snapshot 2026-11-21.