Benchmark Almanac

Cost per Successful Task

Gemini 3.5 Flash $23 leads since July 2026
Not in model cards

Agents solve 40 Terminal-Bench tasks. The benchmark divides spending across all attempts by the number of successful runs.

Arize and Fireworks publish 2,400 runs for ten models. The distinct score includes the cost of failed attempts.

Made by
Arize AI and Fireworks AI
Released
July 2026
Kind
Terminal tasks
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Efficiency
Language
English
In and out
Text, Code → Actions, Code, Text
Interaction
Agentic
Environment
Terminal
Items from
Another benchmark
Refresh
One fixed set
Run by
Anyone
Based on
Terminal-Bench
In views
Agents
Strongest model tested
GPT-5.5 #21 on the capability index
Last model added
July 2026
Scores refreshed
November 21st, 2026
10
Models scored on its own leaderboard

Gemini 3.5 Flash leads with $23 among models released by July 2026

Best score per model from the benchmark's own leaderboard, read on November 21st, 2026. Where a model ran with several settings or harnesses, we keep its best.

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

  1. 1 Gemini 3.5 Flash $23
  2. 2 gpt-oss-120b $33
  3. 3 DeepSeek-V4-Pro $39
  4. 4 Gemini 3.1 Flash-Lite $40
  5. 5 GPT-5 $41
  6. 6 Kimi K2.6 $42
  7. 7 GLM-5.2 $42
  8. 8 Claude Sonnet 5 $49
  9. 9 Kimi K3 $66
  10. 10 GPT-5.5 $67

Source: Official Cost per Successful Task leaderboard

#ModelScoreSettingDateSource
1 Gemini 3.5 Flash $23 May 2026 Official leaderboard
2 gpt-oss-120b $33 Aug 2025 Official leaderboard
3 DeepSeek-V4-Pro $39 Apr 2026 Official leaderboard
4 Gemini 3.1 Flash-Lite $40 Mar 2026 Official leaderboard
5 GPT-5 $41 Aug 2025 Official leaderboard
6 Kimi K2.6 $42 Apr 2026 Official leaderboard
7 GLM-5.2 $42 Jun 2026 Official leaderboard
8 Claude Sonnet 5 $49 Jun 2026 Official leaderboard
9 Kimi K3 $66 Jul 2026 Official leaderboard
10 GPT-5.5 $67 Apr 2026 Official leaderboard

Pass rate versus cost per successful task

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% $0$0.25$0.5$0.75$1$1.25 Cost per successful task ($), lower is better Pass rate GPT-5.5Kimi K2.6Gemini 3.1 Flash-Litegpt-oss-120b 0% 20% 40% 60% 80% $0$0.25$0.5$0.75$1$1.25 Cost per successful task ($), lower is better Pass rate GPT-5.5Kimi K2.6Gemini 3.1 Flash-Litegpt-oss-120b
ModelSettingPass rateCost per successful task
GPT-5.5 67% $0.64
Kimi K3 66% $0.67
Claude Sonnet 5 49% $1.01
Kimi K2.6 42% $0.38
GLM-5.2 42% $0.5
GPT-5 41% $0.77
Gemini 3.1 Flash-Lite 40% $0.06
DeepSeek-V4-Pro 39% $0.59
gpt-oss-120b 33% $0.05
Gemini 3.5 Flash 23% $1.23

Number one over time

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

Google DeepMind
Own leaderboard Not tested since Jul 2026

Source: Official Cost per Successful Task leaderboard

Scores over time

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

Plot with other benchmarks
On the benchmark’s own leaderboard
$20 $30 $40 $50 $60 $70 2026 $20 $30 $40 $50 $60 $70 2026

Source: Official Cost per Successful Task leaderboard

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 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (10 entries).
  • Pass Models scored 10 models we track.
  • Pass Newest score The newest is from July 2026, for Kimi K3.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
  • Note Headroom Scored as US dollars, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arize.com: 10 rows for models we track at the last read.
  • Pass Last read November 21st, 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