Benchmark leaderboard
Terminal-Bench 4.0 is the official leaderboard of the terminal-work benchmark. Every published row measures a model through a vendor's own agent such as Claude Code, Codex or Grok Build, so ModelCap shows each result with its harness named and does not score the Index on it. This page lists every model ModelCap tracks with a published Terminal-Bench 4.0 result, its best evaluated configuration, the source's own rank, and the model's ModelCap Index position, API price and context window.
Snapshot as of 29 September 2026
Claude Opus 5.5 from Anthropic leads the Terminal-Bench 4.0 leaderboard among the 23 tracked models with 64.9% (Claude Code · Max configuration), per the source snapshot published 29 September 2026.
Sorted by published score, one row per model. Source: Terminal-Bench (Laude Institute).
Terminal-Bench 4.0 ranks models by task accuracy through the vendor's own agent. ModelCap ingests the board as published, matches each entry to a catalogue model with a reviewed identity, and shows the source's score, interval and rank unchanged. Terminal-Bench 4.0 is displayed as a public reference board and does not contribute to the ModelCap Index. Its published measurements and tested configurations remain visible here. The methodology page lists the sources admitted to the Index. Read the methodology and identity rules.
Terminal-Bench 4.0 is the official leaderboard of the terminal-work benchmark. Every published row measures a model through a vendor's own agent such as Claude Code, Codex or Grok Build, so ModelCap shows each result with its harness named and does not score the Index on it. It is published by Terminal-Bench (Laude Institute).
Claude Opus 5.5 (Anthropic) holds the top Terminal-Bench 4.0 score among the models ModelCap tracks, at 64.9% as of the source snapshot published 29 September 2026.
23 tracked models have a published Terminal-Bench 4.0 result on ModelCap; the source board itself lists 35 entries. Each row shows the model's best evaluated configuration.
GLM 5.3 Flash from Z.ai is the highest-scoring model with openly downloadable weights on this board, at 35.8%.
Among the ten highest-scoring priced models, GLM 5.3 has the lowest listed output price at $1.80 per 1M tokens while scoring 41.8%.
The board ranks models by task accuracy through the vendor's own agent; higher scores are better. ModelCap shows the source's own score, interval and rank and never re-runs the evaluation.
Terminal-Bench 4.0 is displayed as a public reference board and does not contribute to the ModelCap Index. Its published measurements and tested configurations remain visible here. The methodology page lists the sources admitted to the Index.
The newest Terminal-Bench 4.0 publication ModelCap holds is dated 29 September 2026. This page uses the published snapshot; source collection, publication and page updates happen separately.