Benchmark Almanac

WorkplaceAgents

Gemini 4 Argon 21.2% leads since October 2026
Not in model cards

Agents produce finished work products for 200 professional tasks across 96 occupations in 19 sectors. Harbor grades deliverables with code checks and rubrics.

Cost per trial ranged from $0.59 to $32.16 across models.

Made by
Snorkel AI
Released
October 2026
Kind
Tasks from real jobs
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
14
Models scored on its own leaderboard

Gemini 4 Argon leads with 21.2%

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

14 models. Scroll sideways for the rest.

  1. 21.2 Gemini 4 Argon
  2. 17.4 Muse Wonder
  3. 17.1 Grok 4.6
  4. 15.8 Claude Fable 5.1
  5. 15.2 Grok 4.7
  6. 15.0 GPT-6.1 Sol
  7. 14.9 Claude Opus 5
  8. 14.4 GPT-6 Astra
  9. 14.4 Claude Opus 5.5
  10. 13.0 Kimi K3
  11. 12.8 GLM-5.3
  12. 11.2 Gemini 3.8 Flash
  13. 8.60 DeepSeek-V4-Pro
  14. 6.60 Nemotron 3 Ultra
  1. 1 Gemini 4 Argon 21.2%
  2. 2 Muse Wonder 17.4%
  3. 3 Grok 4.6 17.1%
  4. 4 Claude Fable 5.1 15.8%
  5. 5 Grok 4.7 15.2%
  6. 6 GPT-6.1 Sol 15%
  7. 7 Claude Opus 5 14.9%
  8. 8 GPT-6 Astra 14.4%
  9. 9 Claude Opus 5.5 14.4%
  10. 10 Kimi K3 13%
  11. 11 GLM-5.3 12.8%
  12. 12 Gemini 3.8 Flash 11.2%
  13. 13 DeepSeek-V4-Pro 8.60%
  14. 14 Nemotron 3 Ultra 6.60%

Source: Official WorkplaceAgents leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Argon 21.2% Sep 2026 Official leaderboard
2 Muse Wonder 17.4% Oct 2026 Official leaderboard
3 Grok 4.6 17.1% Aug 2026 Official leaderboard
4 Claude Fable 5.1 15.8% Sep 2026 Official leaderboard
5 Grok 4.7 15.2% Sep 2026 Official leaderboard
6 GPT-6.1 Sol 15% Sep 2026 Official leaderboard
7 Claude Opus 5 14.9% Jul 2026 Official leaderboard
8 GPT-6 Astra 14.4% Sep 2026 Official leaderboard
9 Claude Opus 5.5 14.4% Sep 2026 Official leaderboard
10 Kimi K3 13% Jul 2026 Official leaderboard
11 GLM-5.3 12.8% Aug 2026 Official leaderboard
12 Gemini 3.8 Flash 11.2% Sep 2026 Official leaderboard
13 DeepSeek-V4-Pro 8.60% Apr 2026 Official leaderboard
14 Nemotron 3 Ultra 6.60% Jun 2026 Official leaderboard

Pass@1 vs. cost per trial

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% 5% 10% 15% 20% 25% $0$10$20$30$40 Cost per trial ($), lower is better Pass@1 Kimi K3: Pass@1 13%, cost per trial $7.79 Claude Opus 5: Pass@1 14.9%, cost per trial $16 Grok 4.6: Pass@1 17.1%, cost per trial $7.84 GLM-5.3: Pass@1 12.8%, cost per trial $14.3 Claude Fable 5.1: Pass@1 15.8%, cost per trial $18.6 GPT-6 Astra: Pass@1 14.4%, cost per trial $22.6 Grok 4.7: Pass@1 15.2%, cost per trial $12.9 Claude Opus 5.5: Pass@1 14.4%, cost per trial $6.35 GPT-6.1 Sol: Pass@1 15%, cost per trial $5.75 Muse Wonder: Pass@1 17.4%, cost per trial $32.2 DeepSeek-V4-Pro: Pass@1 8.6%, cost per trial $1.61 Nemotron 3 Ultra: Pass@1 6.6%, cost per trial $0.59 Gemini 3.8 Flash: Pass@1 11.2%, cost per trial $3.63 Gemini 4 Argon: Pass@1 21.2%, cost per trial $3.64 Gemini 4 ArgonGemini 3.8 FlashDeepSeek-V4-ProNemotron 3 Ultra 0% 5% 10% 15% 20% 25% $0$10$20$30$40 Cost per trial ($), lower is better Pass@1 Kimi K3: Pass@1 13%, cost per trial $7.79 Claude Opus 5: Pass@1 14.9%, cost per trial $16 Grok 4.6: Pass@1 17.1%, cost per trial $7.84 GLM-5.3: Pass@1 12.8%, cost per trial $14.3 Claude Fable 5.1: Pass@1 15.8%, cost per trial $18.6 GPT-6 Astra: Pass@1 14.4%, cost per trial $22.6 Grok 4.7: Pass@1 15.2%, cost per trial $12.9 Claude Opus 5.5: Pass@1 14.4%, cost per trial $6.35 GPT-6.1 Sol: Pass@1 15%, cost per trial $5.75 Muse Wonder: Pass@1 17.4%, cost per trial $32.2 DeepSeek-V4-Pro: Pass@1 8.6%, cost per trial $1.61 Nemotron 3 Ultra: Pass@1 6.6%, cost per trial $0.59 Gemini 3.8 Flash: Pass@1 11.2%, cost per trial $3.63 Gemini 4 Argon: Pass@1 21.2%, cost per trial $3.64 Gemini 4 ArgonGemini 3.8 FlashDeepSeek-V4-ProNemotron 3 Ultra
ModelSettingPass@1Cost per trial
Gemini 4 Argon 21.2% $3.64
Muse Wonder 17.4% $32.16
Grok 4.6 17.1% $7.84
Claude Fable 5.1 15.8% $18.65
Grok 4.7 15.2% $12.86
GPT-6.1 Sol 15% $5.75
Claude Opus 5 14.9% $15.96
Claude Opus 5.5 14.4% $6.35
GPT-6 Astra 14.4% $22.6
Kimi K3 13% $7.79
GLM-5.3 12.8% $14.33
Gemini 3.8 Flash 11.2% $3.63
DeepSeek-V4-Pro 8.6% $1.61
Nemotron 3 Ultra 6.6% $0.59

The best score, 21.2%, costs $3.64 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek NVIDIA Google DeepMind Cost frontier One model at each effort level
0% 5% 10% 15% 20% 25% $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Argon Gemini 3.8 Flash DeepSeek-V4-Pro Nemotron 3 Ultra 0% 5% 10% 15% 20% 25% $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Argon Gemini 3.8 Flash

Source: Official WorkplaceAgents leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon 21.2% $3.64
Muse Wonder 17.4% $32
Grok 4.6 17.1% $7.84
Claude Fable 5.1 15.8% $19
Grok 4.7 15.2% $13
GPT-6.1 Sol 15% $5.75
Claude Opus 5 14.9% $16
GPT-6 Astra 14.4% $23
Claude Opus 5.5 14.4% $6.35
Kimi K3 13% $7.79
GLM-5.3 12.8% $14
Gemini 3.8 Flash 11.2% $3.63
DeepSeek-V4-Pro 8.60% $1.61
Nemotron 3 Ultra 6.60% $0.59

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 1 leader
Gemini 4 Argon 21.2%

Source: Official WorkplaceAgents 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%) Gemini 4 Argon Claude Opus 5 DeepSeek-V4-Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Claude Opus 5

Source: Official WorkplaceAgents 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 Economic work

All 97 in Economic work
  • JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
  • GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate. Seed2.1 Pro 87.9%
  • GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
  • AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
  • CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%
  • OfficeQA In use Grounded reasoning over a large corpus of U.S. Treasury Bulletin documents: locate tables and compute precise numeric answers. Claude Opus 4.7 86.3%