Benchmark Almanac

DAYJOB: Accounting

Claude Sonnet 5.5 35.2% leads since October 2026
Not in model cards

Agents do accounting assignments from a short request and a workspace of files. An agent judge applies an expert rubric, and a pass needs 95% of criteria.

Results for 15 models appear only in the Fireworks Specialized Intelligence Index. Surge had not published the tasks or task count by October 30th, 2026.

Made by
Surge AI
Released
October 2026
Kind
Finance
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
17
Models scored on its own leaderboard

Claude Sonnet 5.5 leads with 35.2%

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.

17 models. Scroll sideways for the rest.

  1. 35.2 Claude Sonnet 5.5
  2. 34.4 Claude Opus 5.5
  3. 30.8 Muse Wonder
  4. 30.0 GPT-6 Astra
  5. 27.2 Claude Fable 5.1
  6. 25.2 GPT-6.1 Sol
  7. 20.0 Grok 4.7
  8. 18.0 Muse Spark 1.3
  9. 18.0 Gemini 4 Argon
  10. 13.6 Kimi K3.1 Pro
  11. 8.40 Tencent Hy4 preview
  12. 7.60 GLM-5.3
  13. 7.60 DeepSeek V4.1 Flash 1017
  14. 6.00 Kimi K3
  15. 6.00 Gemini 3.8 Flash
  16. 4.40 Qwen 3.8 Max
  17. 3.60 Mistral Large 4
  1. 1 Claude Sonnet 5.5 adaptive/max 35.2%
  2. 2 Claude Opus 5.5 adaptive/max 34.4%
  3. 3 Muse Wonder max 30.8%
  4. 4 GPT-6 Astra max 30%
  5. 5 Claude Fable 5.1 adaptive/max 27.2%
  6. 6 GPT-6.1 Sol max 25.2%
  7. 7 Grok 4.7 xhigh 20%
  8. 8 Muse Spark 1.3 max 18%
  9. 9 Gemini 4 Argon high 18%
  10. 10 Kimi K3.1 Pro 13.6%
  11. 11 Tencent Hy4 preview high 8.40%
  12. 12 GLM-5.3 max 7.60%
  13. 13 DeepSeek V4.1 Flash 1017 max 7.60%
  14. 14 Kimi K3 max 6.00%
  15. 15 Gemini 3.8 Flash high 6.00%
  16. 16 Qwen 3.8 Max xhigh 4.40%
  17. 17 Mistral Large 4 high 3.60%

Source: Official DAYJOB: Accounting leaderboard

#ModelScoreSettingDateSource
1 Claude Sonnet 5.5 35.2% adaptive/max Sep 2026 Official leaderboard
2 Claude Opus 5.5 34.4% adaptive/max Sep 2026 Official leaderboard
3 Muse Wonder 30.8% max Oct 2026 Official leaderboard
4 GPT-6 Astra 30% max Sep 2026 Official leaderboard
5 Claude Fable 5.1 27.2% adaptive/max Sep 2026 Official leaderboard
6 GPT-6.1 Sol 25.2% max Sep 2026 Official leaderboard
7 Grok 4.7 20% xhigh Sep 2026 Official leaderboard
8 Muse Spark 1.3 18% max Sep 2026 Official leaderboard
9 Gemini 4 Argon 18% high Sep 2026 Official leaderboard
10 Kimi K3.1 Pro 13.6% Oct 2026 Official leaderboard
11 Tencent Hy4 preview 8.40% high Aug 2026 Official leaderboard
12 GLM-5.3 7.60% max Aug 2026 Official leaderboard
13 DeepSeek V4.1 Flash 1017 7.60% max Oct 2026 Official leaderboard
14 Kimi K3 6.00% max Jul 2026 Official leaderboard
15 Gemini 3.8 Flash 6.00% high Sep 2026 Official leaderboard
16 Qwen 3.8 Max 4.40% xhigh Aug 2026 Official leaderboard
17 Mistral Large 4 3.60% high Oct 2026 Official leaderboard

Accounting score versus cost

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% 10% 20% 30% 40% $0$5$10$15$20 Model cost per task at board pricing ($), lower is better Pass@1 at 95% Kimi K3 (max): Pass@1 at 95% 6%, model cost per task at board pricing $3.32 Qwen 3.8 Max (xhigh): Pass@1 at 95% 4.4%, model cost per task at board pricing $4.28 GLM-5.3 (max): Pass@1 at 95% 7.6%, model cost per task at board pricing $6.63 Claude Fable 5.1 (adaptive/max): Pass@1 at 95% 27.2%, model cost per task at board pricing $15.1 Gemini 3.8 Flash (high): Pass@1 at 95% 6%, model cost per task at board pricing $1.87 Muse Spark 1.3 (max): Pass@1 at 95% 18%, model cost per task at board pricing $1.97 GPT-6 Astra (max): Pass@1 at 95% 30%, model cost per task at board pricing $10.3 Grok 4.7 (xhigh): Pass@1 at 95% 20%, model cost per task at board pricing $5.63 Claude Opus 5.5 (adaptive/max): Pass@1 at 95% 34.4%, model cost per task at board pricing $11.7 Gemini 4 Argon (high): Pass@1 at 95% 18%, model cost per task at board pricing $3.84 Mistral Large 4 (high): Pass@1 at 95% 3.6%, model cost per task at board pricing $4.19 Kimi K3.1 Pro: Pass@1 at 95% 13.6%, model cost per task at board pricing $3.47 Tencent Hy4 preview (high): Pass@1 at 95% 8.4%, model cost per task at board pricing $1 Claude Sonnet 5.5 (adaptive/max): Pass@1 at 95% 35.2%, model cost per task at board pricing $10.2 GPT-6.1 Sol (max): Pass@1 at 95% 25.2%, model cost per task at board pricing $1.54 Muse Wonder (max): Pass@1 at 95% 30.8%, model cost per task at board pricing $6.64 DeepSeek V4.1 Flash 1017 (max): Pass@1 at 95% 7.6%, model cost per task at board pricing $0.54 Claude Sonnet 5.5 (adaptive/max)Muse Wonder (max)GPT-6.1 Sol (max)Tencent Hy4 preview (high)DeepSeek V4.1 Flash 1017 (max) 0% 10% 20% 30% 40% $0$5$10$15$20 Model cost per task at board pricing ($), lower is better Pass@1 at 95% Kimi K3 (max): Pass@1 at 95% 6%, model cost per task at board pricing $3.32 Qwen 3.8 Max (xhigh): Pass@1 at 95% 4.4%, model cost per task at board pricing $4.28 GLM-5.3 (max): Pass@1 at 95% 7.6%, model cost per task at board pricing $6.63 Claude Fable 5.1 (adaptive/max): Pass@1 at 95% 27.2%, model cost per task at board pricing $15.1 Gemini 3.8 Flash (high): Pass@1 at 95% 6%, model cost per task at board pricing $1.87 Muse Spark 1.3 (max): Pass@1 at 95% 18%, model cost per task at board pricing $1.97 GPT-6 Astra (max): Pass@1 at 95% 30%, model cost per task at board pricing $10.3 Grok 4.7 (xhigh): Pass@1 at 95% 20%, model cost per task at board pricing $5.63 Claude Opus 5.5 (adaptive/max): Pass@1 at 95% 34.4%, model cost per task at board pricing $11.7 Gemini 4 Argon (high): Pass@1 at 95% 18%, model cost per task at board pricing $3.84 Mistral Large 4 (high): Pass@1 at 95% 3.6%, model cost per task at board pricing $4.19 Kimi K3.1 Pro: Pass@1 at 95% 13.6%, model cost per task at board pricing $3.47 Tencent Hy4 preview (high): Pass@1 at 95% 8.4%, model cost per task at board pricing $1 Claude Sonnet 5.5 (adaptive/max): Pass@1 at 95% 35.2%, model cost per task at board pricing $10.2 GPT-6.1 Sol (max): Pass@1 at 95% 25.2%, model cost per task at board pricing $1.54 Muse Wonder (max): Pass@1 at 95% 30.8%, model cost per task at board pricing $6.64 DeepSeek V4.1 Flash 1017 (max): Pass@1 at 95% 7.6%, model cost per task at board pricing $0.54 Claude Sonnet 5.5 (adaptive/max)Muse Wonder (max)GPT-6.1 Sol (max)Tencent Hy4 preview (high)DeepSeek V4.1 Flash 1017 (max)
ModelSettingPass@1 at 95%Model cost per task at board pricing
Claude Sonnet 5.5 adaptive/max 35.2% $10.24
Claude Opus 5.5 adaptive/max 34.4% $11.66
Muse Wonder max 30.8% $6.64
GPT-6 Astra max 30% $10.27
Claude Fable 5.1 adaptive/max 27.2% $15.12
GPT-6.1 Sol max 25.2% $1.54
Grok 4.7 xhigh 20% $5.63
Muse Spark 1.3 max 18% $1.97
Gemini 4 Argon high 18% $3.84
Kimi K3.1 Pro 13.6% $3.47
Tencent Hy4 preview high 8.4% $1
DeepSeek V4.1 Flash 1017 max 7.6% $0.54
GLM-5.3 max 7.6% $6.63
Gemini 3.8 Flash high 6% $1.87
Kimi K3 max 6% $3.32
Qwen 3.8 Max xhigh 4.4% $4.28
Mistral Large 4 high 3.6% $4.19

The best score, 35.2%, costs $10 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.

Tencent Anthropic OpenAI Meta DeepSeek Cost frontier One model at each effort level
0% 10% 20% 30% 40% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Sonnet 5.5, adaptive/max Muse Wonder, max GPT-6.1 Sol, max Tencent Hy4 preview, high 0% 10% 20% 30% 40% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Sonnet 5.5, adaptive/max Muse Wonder, max

Source: Official DAYJOB: Accounting leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Sonnet 5.5 adaptive/max 35.2% $10 436,098
Claude Opus 5.5 adaptive/max 34.4% $12 264,746
Muse Wonder max 30.8% $6.64 115,205
GPT-6 Astra max 30% $10 74,921
Claude Fable 5.1 adaptive/max 27.2% $15 146,702
GPT-6.1 Sol max 25.2% $1.54 81,902
Grok 4.7 xhigh 20% $5.63 149,118
Muse Spark 1.3 max 18% $1.97 96,820
Gemini 4 Argon high 18% $3.84 168,693
Kimi K3.1 Pro 13.6% $3.47
Tencent Hy4 preview high 8.40% $1.00 110,346
GLM-5.3 max 7.60% $6.63 148,215
DeepSeek V4.1 Flash 1017 max 7.60% $0.54 180,923
Kimi K3 max 6.00% $3.32 60,782
Gemini 3.8 Flash high 6.00% $1.87 121,787
Qwen 3.8 Max xhigh 4.40% $4.28 119,540
Mistral Large 4 high 3.60% $4.19 152,269

Number one over time

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

Anthropic
Own leaderboard 1 leader
Claude Sonnet 5.5 35.2%

Source: Official DAYJOB: Accounting 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% 2027 Near the ceiling (85%) Claude Sonnet 5.5 GPT-6 Astra Tencent Hy4 preview Kimi K3 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Claude Sonnet 5.5 Claude Fable 5.1

Source: Official DAYJOB: Accounting 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 Law, finance and medicine

All 179 in Law, finance and medicine
  • MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy. Qwen 3.8 Max 80.4%
  • HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
  • HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy. Ling-3.1-flash 81.7%
  • KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items. Solar Pro 4 77.5%
  • KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. Solar Pro 4 93.2%