Benchmark Almanac

OpenThoughts-TBLite

Kimi K2.5 75.1% leads since February 2026
Not in model cards

Agents solve 100 Terminal-Bench tasks picked to track Terminal-Bench 2 scores while running faster. Tasks are balanced across four difficulty bands using Claude Haiku 4.5 pass rates.

Built as a development set for smaller models. It replaces the earlier OpenThoughts-TB-dev set. Runs use the Harbor framework.

Released
February 2026
Kind
Terminal tasks
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Strongest model tested
Kimi K2.5 #75 on the capability index
Last model added
January 2026
Scores refreshed
November 12th, 2026
7
Entries on its own leaderboard, 4 models we track

Kimi K2.5 leads with 75.1% among models released by January 2026

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

The newest model with a score here came out in January 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 75.1 Kimi K2.5
  2. 67.7 GLM-4.7
  3. 64.4 Claude Haiku 4.5
  4. 50.5 GPT-5 mini
  5. 42.1 Qwen3-Coder-480B-A35B-Instruct-FP8
  6. 37.0 Qwen3-235B-A22B-Instruct-2507-tput
  7. 21.5 NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
  1. 1 Kimi K2.5 75.1%
  2. 2 GLM-4.7 67.7%
  3. 3 Claude Haiku 4.5 64.4%
  4. 4 GPT-5 mini 50.5%
  5. 5 Qwen3-Coder-480B-A35B-Instruct-FP8 42.1%
  6. 6 Qwen3-235B-A22B-Instruct-2507-tput 37%
  7. 7 NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 21.5%

Source: Official OpenThoughts-TBLite leaderboard

#EntryScoreSettingDateSource
1 Kimi K2.5 75.1% Jan 2026 Official leaderboard
2 GLM-4.7 67.7% Dec 2025 Official leaderboard
3 Claude Haiku 4.5 64.4% Oct 2025 Official leaderboard
4 GPT-5 mini 50.5% Aug 2025 Official leaderboard
5 Qwen3-Coder-480B-A35B-Instruct-FP8 42.1% Official leaderboard
6 Qwen3-235B-A22B-Instruct-2507-tput 37% Official leaderboard
7 NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 21.5% Official leaderboard

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Moonshot AI
Own leaderboard Not tested since Feb 2026
Kimi K2.5 75.1%

Source: Official OpenThoughts-TBLite leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Kimi K2.5 GLM-4.7 Claude Haiku 4.5 GPT-5 mini 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Kimi K2.5 GLM-4.7

Source: Official OpenThoughts-TBLite leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Terminal agents

All 19 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 67.3%
  • 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%
  • 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 4.1 Not in model cards 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. Claude Opus 5.5 65.8%
  • o11y-bench Not in model cards Agents query metrics, logs and traces in a live Grafana stack, investigate incidents and fix dashboards. Grading checks claims against facts read from the same environment. Claude Opus 5.5 93.8%