Benchmark Almanac

harness-bench-fast

free-code 2.1.119 + Claude Opus 4.7 100% tops its leaderboard
Not in model cards

Agents operate on files and code using tools and memory. Harbor exports 231 tasks from version 0.3.0. Later public leaderboards report Light and Medium suites separately.

On November 19, 2026, DeepSeek V4.1 Flash 1017 Uncensored with Codex led Medium90 at 78 of 90. This score uses the newer suite.

Made by
AI Forever
Released
May 2026
Kind
Terminal tasks
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Code → Actions, Code
Interaction
Agentic
Environment
Terminal
Items from
Another benchmark
Run by
Anyone
In views
Agents
13
Entries on its own leaderboard, 0 models we track

free-code 2.1.119 + Claude Opus 4.7 tops the leaderboard with 100% No scores yet

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. Agents built around a model, such as free-code 2.1.119 + Claude Opus 4.7, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.

  1. 1 free-code 2.1.119 + Claude Opus 4.7 v0.3.0, Harness adaptation: yes (built-in + AGENTS.md inject) 100%
  2. 2 free-code 2.1.119 + Claude Haiku 4.5 v0.3.0, Harness adaptation: yes (built-in + AGENTS.md inject) 96.1%
  3. 3 ouroboros + Claude Sonnet 4.6 v0.3.0, via OpenRouter, native tool calls, Harness adaptation: yes (Ouroboros CLI adapter) 96.1%
  4. 4 ouroboros + Claude Haiku 4.5 v0.3.0, via OpenRouter, native tool calls, Harness adaptation: yes (Ouroboros CLI adapter) 93.1%
  5. 5 deepagents + Claude Haiku 4.5 v0.3.0, via OpenRouter, max_tokens=4096, Harness adaptation: no 90.5%
  6. 6 deepagents + MiniMax-M2 v0.3.0, via OpenRouter, Harness adaptation: no 90.5%
  7. 7 deepagents + DeepSeek-V3.2-Exp v0.3.0, via OpenRouter, Harness adaptation: no 90%
  8. 8 deepagents + GLM-4.6 v0.3.0, via OpenRouter, Harness adaptation: no 89.2%
  9. 9 GigaChat-3-Ultra v0.3.0, PROM, deepagents 0.6.3 + langgraph 1.2.1, Harness adaptation: yes (v9 + memory wiring) 84.4%
  10. 10 deepagents + DeepSeek-V4-Flash v0.3.0, 284B-A13B MoE, Harness adaptation: no 80.5%
  11. 11 deepagents + gpt-oss-120b v0.3.0, 120B dense, Harness adaptation: no 71.4%
  12. 12 deepagents + Qwen3-235B-A22B-Instruct Jul 2025, v0.3.0, Harness adaptation: no 70.1%
  13. 13 GLM-4-32B v0.3.0, 32B dense, Harness adaptation: no 32.9%

Source: Official harness-bench-fast leaderboard

#EntryScoreSettingReleasedRun onSource
1 free-code 2.1.119 + Claude Opus 4.7 100% v0.3.0, Harness adaptation: yes (built-in + AGENTS.md inject) Apr 2026 May 2026 Official leaderboard
2 free-code 2.1.119 + Claude Haiku 4.5 96.1% v0.3.0, Harness adaptation: yes (built-in + AGENTS.md inject) Oct 2025 May 2026 Official leaderboard
3 ouroboros + Claude Sonnet 4.6 96.1% v0.3.0, via OpenRouter, native tool calls, Harness adaptation: yes (Ouroboros CLI adapter) Feb 2026 May 2026 Official leaderboard
4 ouroboros + Claude Haiku 4.5 93.1% v0.3.0, via OpenRouter, native tool calls, Harness adaptation: yes (Ouroboros CLI adapter) Oct 2025 May 2026 Official leaderboard
5 deepagents + Claude Haiku 4.5 90.5% v0.3.0, via OpenRouter, max_tokens=4096, Harness adaptation: no Oct 2025 May 2026 Official leaderboard
6 deepagents + MiniMax-M2 90.5% v0.3.0, via OpenRouter, Harness adaptation: no Oct 2025 May 2026 Official leaderboard
7 deepagents + DeepSeek-V3.2-Exp 90% v0.3.0, via OpenRouter, Harness adaptation: no Sep 2025 May 2026 Official leaderboard
8 deepagents + GLM-4.6 89.2% v0.3.0, via OpenRouter, Harness adaptation: no Sep 2025 May 2026 Official leaderboard
9 GigaChat-3-Ultra 84.4% v0.3.0, PROM, deepagents 0.6.3 + langgraph 1.2.1, Harness adaptation: yes (v9 + memory wiring), deepagents May 2026 Official leaderboard
10 deepagents + DeepSeek-V4-Flash 80.5% v0.3.0, 284B-A13B MoE, Harness adaptation: no Apr 2026 May 2026 Official leaderboard
11 deepagents + gpt-oss-120b 71.4% v0.3.0, 120B dense, Harness adaptation: no Aug 2025 May 2026 Official leaderboard
12 deepagents + Qwen3-235B-A22B-Instruct Jul 2025 70.1% v0.3.0, Harness adaptation: no Jul 2025 May 2026 Official leaderboard
13 GLM-4-32B 32.9% v0.3.0, 32B dense, Harness adaptation: no, deepagents May 2026 Official leaderboard
#ModelScoreSettingReleasedRun onSource

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Terminal agents

All 35 in Terminal agents
  • SkillsBench In use 86-87 containerized expert tasks in 11 domains, run with and without Agent Skills. Verifier tests give pass rate and skill lift. GPT-5.5 66.5%
  • Terminal-Bench Hard In use Agents complete 44 hard tasks from Terminal-Bench Core in the Terminus 2 harness. Artificial Analysis scores pass@1 averaged over 3 repeats. Gemma 4 31B IT 36%
  • Terminal-Bench 4.1 In use The Terminal-Bench 4.0 tasks with hardened verifiers: agents complete 66 hard tasks in containerized terminals across software, ML, science, security, and operations. Test scripts verify each result, scored as resolution rate. Copilot Fusion 71.2%
  • Harbor-Index In use Agents solve about 80 hard tasks drawn from 29 benchmarks run through Harbor adapters, such as HLE, GSO and SWE-bench Pro. Each task keeps its source verifier. GPT-5.5 28.1%
  • Terminal-Bench 2.1 Extended Not in model cards Agents complete long command-line tasks in containers, scored by verification tests. Mercor adds 99 new expert-written tasks to the 89 public ones and reports both scores. Muse Wonder 54%
  • Terminal-Bench Pro Not in model cards Agents solve 400 terminal tasks across eight domains. Harbor distributes the 200 public tasks. Alibaba keeps the other 200 for private evaluation. Claude Opus 4.5 51%

Data audit

All 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the official leaderboard (13 entries).
  • Pass Models scored 13 entries on its leaderboard, none a model we track.
  • Note Newest score No scores yet.
  • Note Frontier models No models scored yet.
  • Note Headroom No scores yet.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 11 rows for models we track and 2 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check