Benchmark Almanac

LevelBench-100 v4.1

Level Playbooks (2,200+ Past Experiences) + DeepSeek-V3 Mar 2025 100% tops its leaderboard GPT-6.1 Sol 81% leads as a model on its own Gemini 4 Argon 71% leads in a standard harness
Not in model cards
Version 4.1

Models take over the books of small businesses using QuickBooks exports. They reconcile balances, identify missing evidence, and prepare closing entries across 12 workflows.

Level publishes 100 binary checks and 38 configurations. The 27 unassisted rows cover 22 distinct models after combining reasoning settings.

Made by
Level
Released
October 2026
Kind
Finance
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Tables → Actions, Text, Tables
Interaction
Agentic
Environment
Browser
Items from
Real work and use
Run by
Anyone
Occupations
Management, Business and financial operations
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
October 2026
Scores refreshed
November 20th, 2026
31
Entries on its own leaderboard, 22 models we track

Level Playbooks (2,200+ Past Experiences) + DeepSeek-V3 tops the leaderboard with 100% GPT-6.1 Sol leads with 81% Gemini 4 Argon leads with 71% in a standard harness

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. The best model on its own is GPT-6.1 Sol, at 81%. Agents built around a model, such as Level Playbooks (2,200+ Past Experiences) + DeepSeek-V3, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best. This view keeps only runs in a standard harness, a harness no model’s maker built. A maker’s own agent, such as Claude Code with a Claude model, is left out.

Source: Official LevelBench-100 v4.1 leaderboard

#EntryScoreSettingDateSource
1 Level Playbooks (2,200+ Past Experiences) + DeepSeek-V3 Mar 2025 100% Mar 2025 Official leaderboard
2 Level Playbooks (2,200+ Past Experiences) + GPT-5.6 Terra 100% Jul 2026 Official leaderboard
3 Level Playbooks (2,200+ Past Experiences) + Gemini 3.8 Flash 100% Sep 2026 Official leaderboard
4 Level Playbooks (2,200+ Past Experiences) + Claude Opus 5.5 100% Sep 2026 Official leaderboard
5 Level Playbooks (2,200+ Past Experiences) + GPT-6.1 Sol 100% Sep 2026 Official leaderboard
6 Level Playbooks (2,200+ Past Experiences) + Gemini 4 Argon 100% Sep 2026 Official leaderboard
7 Level Playbooks (2,200+ Past Experiences) + Claude Haiku 5.5 100% Oct 2026 Official leaderboard
8 Gemini CLI 0.64.0 + Gemini 4 Argon 84% Medium Sep 2026 Official leaderboard
9 Antigravity CLI 1.4.0 + Gemini 4 Argon 83% Medium Sep 2026 Official leaderboard
10 GPT-6.1 Sol 81% Codex CLI Sep 2026 Official leaderboard
11 GPT-5.6 Terra 76% Codex CLI Jul 2026 Official leaderboard
12 Gemini 4 Argon 71% Medium Thinking Sep 2026 Official leaderboard
13 Claude Opus 5.5 69% Sep 2026 Official leaderboard
14 Claude Fable 5.1 68% Sep 2026 Official leaderboard
15 Claude Sonnet 5.5 64% Sep 2026 Official leaderboard
16 GPT-6 Astra 60% Medium Effort Sep 2026 Official leaderboard
17 Gemini 3.8 Flash 58% Sep 2026 Official leaderboard
18 Claude Haiku 5.5 55% Oct 2026 Official leaderboard
19 Claude Opus 4.7 54% Apr 2026 Official leaderboard
20 GPT-5.6 Sol 53% Jul 2026 Official leaderboard
21 Gemini 3.1 Pro 52% Feb 2026 Official leaderboard
22 Grok 4.5 50% Jul 2026 Official leaderboard
23 GPT-6 Luna 49% Medium Effort Sep 2026 Official leaderboard
24 Gemini 3.5 Flash 48% May 2026 Official leaderboard
25 DeepSeek-V3 Mar 2025 46% Mar 2025 Official leaderboard
26 Grok 4.3 43% Apr 2026 Official leaderboard
27 Kimi K2.5 41% Jan 2026 Official leaderboard
28 Kimi K2 Jul 2025 39% Jul 2025 Official leaderboard
29 Claude Haiku 4.5 38% Oct 2025 Official leaderboard
30 Qwen3-Next-80B-A3B 35% Sep 2025 Official leaderboard
31 Qwen2.5-72B 31% Sep 2024 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6.1 Sol 81% Codex CLI Sep 2026 Official leaderboard
2 GPT-5.6 Terra 76% Codex CLI Jul 2026 Official leaderboard
3 Gemini 4 Argon 71% Medium Thinking Sep 2026 Official leaderboard
4 Claude Opus 5.5 69% Sep 2026 Official leaderboard
5 Claude Fable 5.1 68% Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 64% Sep 2026 Official leaderboard
7 GPT-6 Astra 60% Medium Effort Sep 2026 Official leaderboard
8 Gemini 3.8 Flash 58% Sep 2026 Official leaderboard
9 Claude Haiku 5.5 55% Oct 2026 Official leaderboard
10 Claude Opus 4.7 54% Apr 2026 Official leaderboard
11 GPT-5.6 Sol 53% Jul 2026 Official leaderboard
12 Gemini 3.1 Pro 52% Feb 2026 Official leaderboard
13 Grok 4.5 50% Jul 2026 Official leaderboard
14 GPT-6 Luna 49% Medium Effort Sep 2026 Official leaderboard
15 Gemini 3.5 Flash 48% May 2026 Official leaderboard
16 DeepSeek-V3 Mar 2025 46% Mar 2025 Official leaderboard
17 Grok 4.3 43% Apr 2026 Official leaderboard
18 Kimi K2.5 41% Jan 2026 Official leaderboard
19 Kimi K2 Jul 2025 39% Jul 2025 Official leaderboard
20 Claude Haiku 4.5 38% Oct 2025 Official leaderboard
21 Qwen3-Next-80B-A3B 35% Sep 2025 Official leaderboard
22 Qwen2.5-72B 31% Sep 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 4 Argon 71% Medium Thinking Sep 2026 Official leaderboard
2 Claude Opus 5.5 69% Sep 2026 Official leaderboard
3 Claude Fable 5.1 68% Sep 2026 Official leaderboard
4 GPT-6.1 Sol 67% High Effort Sep 2026 Official leaderboard
5 Claude Sonnet 5.5 64% Sep 2026 Official leaderboard
6 GPT-6 Astra 60% Medium Effort Sep 2026 Official leaderboard
7 Gemini 3.8 Flash 58% Sep 2026 Official leaderboard
8 GPT-5.6 Terra 57% Low Effort Jul 2026 Official leaderboard
9 Claude Haiku 5.5 55% Oct 2026 Official leaderboard
10 Claude Opus 4.7 54% Apr 2026 Official leaderboard
11 GPT-5.6 Sol 53% Jul 2026 Official leaderboard
12 Gemini 3.1 Pro 52% Feb 2026 Official leaderboard
13 Grok 4.5 50% Jul 2026 Official leaderboard
14 GPT-6 Luna 49% Medium Effort Sep 2026 Official leaderboard
15 Gemini 3.5 Flash 48% May 2026 Official leaderboard
16 DeepSeek-V3 Mar 2025 46% Mar 2025 Official leaderboard
17 Grok 4.3 43% Apr 2026 Official leaderboard
18 Kimi K2.5 41% Jan 2026 Official leaderboard
19 Kimi K2 Jul 2025 39% Jul 2025 Official leaderboard
20 Claude Haiku 4.5 38% Oct 2025 Official leaderboard
21 Qwen3-Next-80B-A3B 35% Sep 2025 Official leaderboard
22 Qwen2.5-72B 31% Sep 2024 Official leaderboard

Rankings by part

We read 4 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Bookkeeping, Accrual close and CFO analysis. GPT-5.6 Terra leads 2 of them, and GPT-6.1 Sol leads the other 2. Arrows show how far a model moved from its overall rank among the models we track.

Bookkeeping

22 models we track have a score here.

  1. 1 GPT-5.6 Terra ↑1 76.5%
  2. 2 Claude Opus 5.5 ↑2 76.5%
  3. 3 GPT-6.1 Sol ↓2 76.5%
  4. 4 Gemini 4 Argon ↓1 76.5%
  5. 5 Claude Fable 5.1 73.5%
  6. 6 Gemini 3.8 Flash ↑2 64.7%
  7. 7 GPT-6 Astra 64.7%
  8. 8 Claude Sonnet 5.5 ↓2 64.7%
  9. 9 Claude Haiku 5.5 61.8%
  10. 10 Claude Opus 4.7 58.8%

The best score, 100%, costs $0.008 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.

Agents other companies built, in a pale shade of the model’s lab Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Level Playbooks (2,200+ Past Experiences) + DeepSeek-V3 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Level Playbooks (2,200+ Past Experiences) + DeepSeek-V3

Source: Official LevelBench-100 v4.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Level Playbooks (2,200+ Past Experiences) + DeepSeek-V3 Level Playbooks (2,200+ Past Experiences) 100% $0.008
Level Playbooks (2,200+ Past Experiences) + GPT-5.6 Terra Level Playbooks (2,200+ Past Experiences) 100% $0.03
Level Playbooks (2,200+ Past Experiences) + Gemini 3.8 Flash Level Playbooks (2,200+ Past Experiences) 100% $0.01
Level Playbooks (2,200+ Past Experiences) + Claude Opus 5.5 Level Playbooks (2,200+ Past Experiences) 100% $0.09
Level Playbooks (2,200+ Past Experiences) + GPT-6.1 Sol Level Playbooks (2,200+ Past Experiences) 100% $0.06
Level Playbooks (2,200+ Past Experiences) + Gemini 4 Argon Level Playbooks (2,200+ Past Experiences) 100% $0.04
Level Playbooks (2,200+ Past Experiences) + Claude Haiku 5.5 Level Playbooks (2,200+ Past Experiences) 100% $0.02
Gemini CLI 0.64.0 + Gemini 4 Argon Medium Gemini CLI 0.64.0 84% $0.05
Antigravity CLI 1.4.0 + Gemini 4 Argon Medium Antigravity CLI 1.4.0 83% $0.05
GPT-6.1 Sol Codex CLI 81% $0.14
GPT-5.6 Terra Codex CLI 76% $0.04
Gemini 4 Argon Medium Thinking 71% $0.04
Claude Opus 5.5 69% $0.24
Claude Fable 5.1 68% $0.28
GPT-6.1 Sol High Effort 67% $0.26
GPT-6.1 Sol Medium Effort 66% $0.12
Gemini 4 Argon High Thinking 65% $0.16
Claude Sonnet 5.5 64% $0.08
Gemini 4 Argon Low Thinking 63% $0.03
GPT-6.1 Sol Low Effort 62% $0.08
GPT-6 Astra Medium Effort 60% $0.07
Gemini 3.8 Flash 58% $0.01
GPT-5.6 Terra Low Effort 57% $0.03
GPT-5.6 Terra Medium Effort 56% $0.04
Claude Haiku 5.5 55% $0.02
Claude Opus 4.7 54% $0.22
GPT-5.6 Sol 53% $0.10
Gemini 3.1 Pro 52% $0.06
Grok 4.5 50% $0.09
GPT-6 Luna Medium Effort 49% $0.03
Gemini 3.5 Flash 48% $0.01
DeepSeek-V3 Mar 2025 46% $0.008
Grok 4.3 43% $0.03
Kimi K2.5 41% $0.01
Kimi K2 Jul 2025 39% $0.01
Claude Haiku 4.5 38% $0.02
Qwen3-Next-80B-A3B 35% $0.01
Qwen2.5-72B 31% $0.01

The best score, 81%, costs $0.14 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-5.6 Terra comes within 5 points of the best for $0.04 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek Google DeepMind OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-6.1 Sol GPT-5.6 Terra Gemini 4 Argon, Low Thinking Gemini 3.8 Flash Gemini 3.5 Flash 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-6.1 Sol Gemini 4 Argon, Low Thinking

Source: Official LevelBench-100 v4.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.1 Sol Codex CLI 81% $0.14
GPT-5.6 Terra Codex CLI 76% $0.04
Gemini 4 Argon Medium Thinking 71% $0.04
Claude Opus 5.5 69% $0.24
Claude Fable 5.1 68% $0.28
GPT-6.1 Sol High Effort 67% $0.26
GPT-6.1 Sol Medium Effort 66% $0.12
Gemini 4 Argon High Thinking 65% $0.16
Claude Sonnet 5.5 64% $0.08
Gemini 4 Argon Low Thinking 63% $0.03
GPT-6.1 Sol Low Effort 62% $0.08
GPT-6 Astra Medium Effort 60% $0.07
Gemini 3.8 Flash 58% $0.01
GPT-5.6 Terra Low Effort 57% $0.03
GPT-5.6 Terra Medium Effort 56% $0.04
Claude Haiku 5.5 55% $0.02
Claude Opus 4.7 54% $0.22
GPT-5.6 Sol 53% $0.10
Gemini 3.1 Pro 52% $0.06
Grok 4.5 50% $0.09
GPT-6 Luna Medium Effort 49% $0.03
Gemini 3.5 Flash 48% $0.01
DeepSeek-V3 Mar 2025 46% $0.008
Grok 4.3 43% $0.03
Kimi K2.5 41% $0.01
Kimi K2 Jul 2025 39% $0.01
Claude Haiku 4.5 38% $0.02
Qwen3-Next-80B-A3B 35% $0.01
Qwen2.5-72B 31% $0.01

The best score, 71%, costs $0.04 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 Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Gemini 4 Argon, Medium Thinking Gemini 4 Argon, Low Thinking Gemini 3.8 Flash Gemini 3.5 Flash 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Gemini 4 Argon, Medium Thinking Gemini 4 Argon, Low Thinking

Source: Official LevelBench-100 v4.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon Medium Thinking 71% $0.04
Claude Opus 5.5 69% $0.24
Claude Fable 5.1 68% $0.28
GPT-6.1 Sol High Effort 67% $0.26
GPT-6.1 Sol Medium Effort 66% $0.12
Gemini 4 Argon High Thinking 65% $0.16
Claude Sonnet 5.5 64% $0.08
Gemini 4 Argon Low Thinking 63% $0.03
GPT-6.1 Sol Low Effort 62% $0.08
GPT-6 Astra Medium Effort 60% $0.07
Gemini 3.8 Flash 58% $0.01
GPT-5.6 Terra Low Effort 57% $0.03
GPT-5.6 Terra Medium Effort 56% $0.04
Claude Haiku 5.5 55% $0.02
Claude Opus 4.7 54% $0.22
GPT-5.6 Sol 53% $0.10
Gemini 3.1 Pro 52% $0.06
Grok 4.5 50% $0.09
GPT-6 Luna Medium Effort 49% $0.03
Gemini 3.5 Flash 48% $0.01
DeepSeek-V3 Mar 2025 46% $0.008
Grok 4.3 43% $0.03
Kimi K2.5 41% $0.01
Kimi K2 Jul 2025 39% $0.01
Claude Haiku 4.5 38% $0.02
Qwen3-Next-80B-A3B 35% $0.01
Qwen2.5-72B 31% $0.01

Number one over time

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

OpenAI
Own leaderboard 1 leader

Source: Official LevelBench-100 v4.1 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 Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-6.1 Sol GPT-5.6 Terra Claude Opus 4.7 DeepSeek-V3 Qwen2.5-72B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-6.1 Sol Claude Opus 4.7

Source: Official LevelBench-100 v4.1 leaderboard

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 4 Argon GPT-5.6 Terra Gemini 3.1 Pro DeepSeek-V3 Qwen2.5-72B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 4 Argon GPT-5.6 Terra

Source: Official LevelBench-100 v4.1 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.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

OpenAI Google DeepMind
LevelBench-100 v5.0 Passed 85% in Nov 2026, by Gemini 4 Carbon
LevelBench-100 v5.0 Passed 85% in Nov 2026, by Gemini 4 Carbon

Source: Official leaderboards

More in Law, finance and medicine

All 637 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%
  • 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. o1-preview 92.7%
  • Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%

Data audit

All 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (31 entries).
  • Pass Models scored 22 models we track, plus 9 other entries on its leaderboard.
  • Pass Newest score The newest is from October 2026, for Claude Haiku 5.5.
  • Pass Frontier models 6 of the top 10 have a score, led by GPT-6 Astra (#4).
  • Pass Headroom The best is 81%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 38 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