OpenCodePapers

code-generation-on-terminal-bench-3-0

Code IntelligenceCode Generation
Dataset Link
Results over time
Click legend items to toggle metrics. Hover points for model names.
Leaderboard
PaperCodeAccuracyModelNameReleaseDate
Failure Recovery via Belief Updates and World Modeling for Long-Horizon Agents66.6BeliefWeaver (Terminus-2)2026-10-13
Terminal-Bench Leaderboard✓ Link42.7Claude Opus 5 (Max) (mini-SWE-agent)2026-07-24
Terminal-Bench Leaderboard✓ Link34.59GPT-5.6 Sol (Max) (Codex)2026-07-09
Terminal-Bench Leaderboard✓ Link34.05Claude Fable 5 (Max) (Claude Code)2026-06-09
DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression✓ Link30.0DeepSeek-V4.1-Flash Max (DeepSeek Harness (Minimal))2026-09-17
GLM-5.3: Frontier Coding with Emergent Cyber Capabilities✓ Link28.3GLM-5.3 (Claude Code)2026-08-14
Terminal-Bench Leaderboard26.49Grok 4.6 (High) (Grok Build)2026-08-12
Terminal-Bench Leaderboard21.08Claude Opus 4.8 (Max) (Claude Code)2026-05-28
Terminal-Bench Leaderboard✓ Link20.81GPT-5.6 Terra (Max) (Codex)2026-07-09
Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-Refine16.95Terminal-RSI-27B2026-10-15
Terminal-Bench Leaderboard15.68Grok 4.5 (XHigh) (Cursor CLI)2026-07-08
Introducing Gemini 3.7 Flash✓ Link14.9Gemini 3.7 Flash (MiniSWE-Agent)2026-08-13
Terminal-Bench Leaderboard14.59Claude Sonnet 5 (Max) (Claude Code)2026-06-30
Terminal-Bench Leaderboard✓ Link14.32GPT-5.6 Luna (Max) (Codex)2026-07-09
Terminal-Bench Leaderboard✓ Link4.59GLM-5.2 (Max) (Claude Code)2026-06-16