Benchmark Almanac

FinDesk-Bench

Gemini 4 Carbon 90.4% leads since November 2026
Not in model cards

Agents find Treasury figures and use them to handle client requests across ten business services. A model grades API calls and the records agents leave behind.

Veris ranks 16 models on 160 tasks. Each model attempts every task three times, and a task passes only when every check passes.

Made by
Veris AI
Released
November 2026
Kind
Documents and spreadsheets
Level
University
Scored
Graded by a model
Answers
Agent task
Measures
Capability
Language
English
In and out
Text → Actions, Text
Interaction
Agentic
Environment
APIs and tools
Items from
Real work and use
Refresh
One fixed set
Run by
A host
Strongest model tested
Gemini 4 Carbon #3 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026
16
Models scored on its own leaderboard

Gemini 4 Carbon leads with 90.4%

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 1 Gemini 4 Carbon high 90.4%
  2. 2 Gemini 4 Hydrogen high 82.5%
  3. 3 Gemini 4 Argon high 78.3%
  4. 4 Claude Opus 5.5 high 72.7%
  5. 5 Grok 4.7 high 63.8%
  6. 6 Claude Sonnet 5.5 high 63.1%
  7. 7 GPT-6.1 Sol high 60.6%
  8. 8 DeepSeek V4.1 Flash high 59%
  9. 9 Mistral Large 4 high 50.4%
  10. 10 Kimi K3 high 49.8%
  11. 11 GLM-5.3-Flash high 47.9%
  12. 12 Inkling 2 Small high 47.1%
  13. 13 Gemini 3.8 Flash high 41.3%
  14. 14 Muse Spark Open high 30%
  15. 15 GPT-5.6 Terra high 19%
  16. 16 Nemotron 3.5 Super VL high 10.2%

Source: Official FinDesk-Bench leaderboard

#ModelScore ±SettingDateSource
1 Gemini 4 Carbon 90.4% 86.2 to 94.2 high, Gemini CLI Oct 2026 Official leaderboard
2 Gemini 4 Hydrogen 82.5% 76.9 to 87.7 high, Gemini CLI Nov 2026 Official leaderboard
3 Gemini 4 Argon 78.3% 72.5 to 83.5 high, Gemini CLI Sep 2026 Official leaderboard
4 Claude Opus 5.5 72.7% 66.7 to 78.3 high, Claude Code Sep 2026 Official leaderboard
5 Grok 4.7 63.8% ±6.5 high, Grok CLI Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 63.1% 56.9 to 69 high, Claude Code Sep 2026 Official leaderboard
7 GPT-6.1 Sol 60.6% ±6.3 high, Codex Sep 2026 Official leaderboard
8 DeepSeek V4.1 Flash 59% ±6.3 high, Pi Sep 2026 Official leaderboard
9 Mistral Large 4 50.4% ±6.5 high, Pi Oct 2026 Official leaderboard
10 Kimi K3 49.8% 43.8 to 56.5 high, Pi Jul 2026 Official leaderboard
11 GLM-5.3-Flash 47.9% 41.9 to 54.4 high, Pi Aug 2026 Official leaderboard
12 Inkling 2 Small 47.1% ±6.6 high, Pi Nov 2026 Official leaderboard
13 Gemini 3.8 Flash 41.3% ±6.3 high, Gemini CLI Sep 2026 Official leaderboard
14 Muse Spark Open 30% 24 to 36.5 high, Pi Oct 2026 Official leaderboard
15 GPT-5.6 Terra 19% 14.2 to 24 high, Codex Jul 2026 Official leaderboard
16 Nemotron 3.5 Super VL 10.2% 6.9 to 14 high, Hermes Oct 2026 Official leaderboard

The best score, 90.4%, costs $2.21 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.

Z.ai DeepSeek Anthropic OpenAI Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.10 $0.20 $0.50 $1.00 $2.00 $5.00 $10 Cost per task, log scale Gemini 4 Carbon, high Gemini 4 Argon, high Claude Opus 5.5, high Claude Sonnet 5.5, high DeepSeek V4.1 Flash, high GLM-5.3-Flash, high 0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 Cost per task, log scale Gemini 4 Carbon, high Gemini 4 Argon, high

Source: Official FinDesk-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Carbon high Gemini CLI 90.4% $2.21
Gemini 4 Hydrogen high Gemini CLI 82.5% $2.89
Gemini 4 Argon high Gemini CLI 78.3% $1.76
Claude Opus 5.5 high Claude Code 72.7% $1.12
Grok 4.7 high Grok CLI 63.8% $1.95
Claude Sonnet 5.5 high Claude Code 63.1% $0.60
GPT-6.1 Sol high Codex 60.6% $0.43
DeepSeek V4.1 Flash high Pi 59% $0.18
Mistral Large 4 high Pi 50.4% $0.59
Kimi K3 high Pi 49.8% $0.94
GLM-5.3-Flash high Pi 47.9% $0.11
Inkling 2 Small high Pi 47.1% $0.15
Gemini 3.8 Flash high Gemini CLI 41.3% $1.03
Muse Spark Open high Pi 30% $1.00
GPT-5.6 Terra high Codex 19% $0.60
Nemotron 3.5 Super VL high Hermes 10.2% $0.12

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 Passed 85% in Nov 2026, by Gemini 4 Carbon

Source: Official FinDesk-Bench leaderboard

Gemini 4 Carbon on the official leaderboard passed 85% in October 2026

Models released before it came out already scored that high. 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 Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Gemini 4 Carbon Gemini 4 Argon Claude Opus 5.5 Grok 4.7 DeepSeek V4.1 Flash Kimi K3 GPT-5.6 Terra 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Gemini 4 Carbon Gemini 4 Argon

Source: Official FinDesk-Bench 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 Economic work

All 210 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
  • Agents' Last Exam v1.1 In use Agents complete 122 public long-horizon professional workflow tasks across 55 sub-industries. Deterministic graders check outcomes against hidden references. GPT-6.2 Astra 47.5%
  • 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%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark

Catalog record 1 warning 1 pass
  • Warning Task access We don’t know whether the tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (16 entries).
  • Pass Models scored 16 models we track.
  • Pass Newest score The newest is from November 2026, for Gemini 4 Hydrogen.
  • Pass Frontier models 5 of the top 10 have a score, led by Gemini 4 Carbon (#3).
  • Note Headroom Saturated. The best is 90.4%, past 85% since October 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on veris.ai: 16 rows for models we track 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