Benchmark Almanac

ApprenticeBench

Claude Fable 5.1 72% leads since September 2026
Not in model cards

Agent joins a simulated company and processes 100 accounts-payable bills in Odoo via GUI or API, learning from records and feedback. Cumulative success rate.

Tests computer use plus continual learning. The company is a simulated California home builder, and the agent first studies six months of past bills, a handbook and an Odoo tutorial. Two human testers set a baseline.

Made by
NeoCognition
Released
September 2026
Kind
Web browsing
Level
University
Scored
Checked automatically
Answers
Agent task
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
September 2026
Scores refreshed
November 12th, 2026
18
Models scored on its own leaderboard

Claude Fable 5.1 leads with 72% among models released by September 2026

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.

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

18 models. Scroll sideways for the rest.

  1. 72.0 Claude Fable 5.1
  2. 68.0 GPT-6 Astra
  3. 43.0 Claude Fable 5
  4. 36.0 Claude Opus 5
  5. 26.0 GPT-5.6 Sol
  6. 24.0 Gemini 3.8 Flash
  7. 20.0 GPT-5.5
  8. 19.0 Muse Spark 1.3
  9. 18.0 Kimi K3
  10. 16.0 Claude Sonnet 5
  11. 16.0 GPT-5.6 Terra
  12. 16.0 Gemini 3.7 Flash
  13. 13.0 Grok 4.6
  14. 11.0 GPT-5.4
  15. 7.00 Claude Opus 4.7
  16. 7.00 GPT-5.6 Luna
  17. 5.00 Claude Opus 4.6
  18. 2.00 Claude Sonnet 4.6
  1. 1 Claude Fable 5.1 max 72%
  2. 2 GPT-6 Astra max 68%
  3. 3 Claude Fable 5 xhigh 43%
  4. 4 Claude Opus 5 max 36%
  5. 5 GPT-5.6 Sol max 26%
  6. 6 Gemini 3.8 Flash max 24%
  7. 7 GPT-5.5 xhigh 20%
  8. 8 Muse Spark 1.3 max 19%
  9. 9 Kimi K3 max 18%
  10. 10 Claude Sonnet 5 max 16%
  11. 11 GPT-5.6 Terra max 16%
  12. 12 Gemini 3.7 Flash max 16%
  13. 13 Grok 4.6 max 13%
  14. 14 GPT-5.4 xhigh 11%
  15. 15 Claude Opus 4.7 max 7.00%
  16. 16 GPT-5.6 Luna max 7.00%
  17. 17 Claude Opus 4.6 max 5.00%
  18. 18 Claude Sonnet 4.6 max 2.00%

Source: Official ApprenticeBench leaderboard

#ModelScoreSettingDateSource
1 Claude Fable 5.1 72% max, Claude Code Sep 2026 Official leaderboard
2 GPT-6 Astra 68% max, Codex Sep 2026 Official leaderboard
3 Claude Fable 5 43% xhigh, Claude Code Jun 2026 Official leaderboard
4 Claude Opus 5 36% max, Claude Code Jul 2026 Official leaderboard
5 GPT-5.6 Sol 26% max, Codex Jul 2026 Official leaderboard
6 Gemini 3.8 Flash 24% max, Claude Code Sep 2026 Official leaderboard
7 GPT-5.5 20% xhigh, Codex Apr 2026 Official leaderboard
8 Muse Spark 1.3 19% max, Claude Code Sep 2026 Official leaderboard
9 Kimi K3 18% max, Claude Code Jul 2026 Official leaderboard
10 Claude Sonnet 5 16% max, Claude Code Jun 2026 Official leaderboard
11 GPT-5.6 Terra 16% max, Codex Jul 2026 Official leaderboard
12 Gemini 3.7 Flash 16% max, Claude Code Aug 2026 Official leaderboard
13 Grok 4.6 13% max, Claude Code Aug 2026 Official leaderboard
14 GPT-5.4 11% xhigh, Codex Mar 2026 Official leaderboard
15 Claude Opus 4.7 7.00% max, Claude Code Apr 2026 Official leaderboard
16 GPT-5.6 Luna 7.00% max, Codex Jul 2026 Official leaderboard
17 Claude Opus 4.6 5.00% max, Claude Code Feb 2026 Official leaderboard
18 Claude Sonnet 4.6 2.00% max, Claude Code Feb 2026 Official leaderboard

The best score, 72%, costs $18 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.

OpenAI Google DeepMind Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.1, max GPT-6 Astra, xhigh GPT-5.6 Sol, max Gemini 3.8 Flash, max Gemini 3.7 Flash, max GPT-5.6 Luna, max 0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.1, max GPT-6 Astra, xhigh

Source: Official ApprenticeBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.1 max Claude Code 72% $18
GPT-6 Astra max Codex 68% $21
GPT-6 Astra xhigh Codex 61% $14
Claude Fable 5.1 xhigh Claude Code 59% $14
Claude Fable 5 xhigh Claude Code 43% $33
Claude Opus 5 max Claude Code 36% $20
Claude Fable 5 high Claude Code 35% $28
Claude Fable 5 max Claude Code 34% $29
GPT-5.6 Sol max Codex 26% $8.68
Gemini 3.8 Flash max Claude Code 24% $4.10
Claude Fable 5 medium Claude Code 23% $22
GPT-5.5 xhigh Codex 20% $13
GPT-5.6 Sol high Codex 20% $6.08
Claude Fable 5 low Claude Code 19% $18
GPT-5.6 Sol xhigh Codex 19% $6.62
Muse Spark 1.3 max Claude Code 19% $7.93
Kimi K3 max Claude Code 18% $26
Claude Sonnet 5 max Claude Code 16% $15
GPT-5.6 Terra max Codex 16% $6.15
Gemini 3.7 Flash max Claude Code 16% $3.92
Grok 4.6 max Claude Code 13% $36
GPT-5.4 xhigh Codex 11% $5.95
GPT-5.6 Sol medium Codex 10% $5.10
Claude Opus 4.7 max Claude Code 7.00% $26
GPT-5.6 Luna max Codex 7.00% $0.77
Claude Opus 4.6 max Claude Code 5.00% $24
GPT-5.6 Sol low Codex 4.00% $3.90
Claude Sonnet 4.6 max Claude Code 2.00% $15
GPT-5.6 Sol none Codex 0% $3.09

Number one over time

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

Anthropic
Own leaderboard Not tested since Sep 2026
Claude Fable 5.1 72%

Source: Official ApprenticeBench 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%) Claude Fable 5.1 Claude Fable 5 GPT-5.5 GPT-5.4 Claude Opus 4.6 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5.1 Claude Fable 5

Source: Official ApprenticeBench 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 Computer and browser use

All 69 in Computer and browser use
  • BrowseComp In use Hard-to-find facts that need persistent web browsing. Short verifiable answers graded for exact correctness. Claude Fable 5 93.3%
  • ScreenSpot-Pro In use GUI grounding on high-resolution professional-software screenshots. The model must locate the target UI element. GPT-6 Astra 92.7%
  • AndroidWorld In use Agents complete 116 tasks across 20 real Android apps in an emulator. Each task checks the device's system state to judge success. GPT-6 Astra 100%
  • OSWorld 2.1 In use Agents complete 108 long-horizon workflows across 31 self-hosted websites and desktop apps, averaging 318 tool calls. Scored by binary completion and checkpoint partial credit. Claude Opus 5.5 81.8%
  • WebArena-Verified In use Agents complete 812 audited WebArena tasks on self-hosted websites. Deterministic evaluators check the agent's answer and captured network traces. Muse Spark 1.1 69%
  • MobileWorld In use Agents complete 201 tasks across 20 mobile apps, including tasks that need user clarification or MCP tools. Backend checks verify success. GPT-5 55.3%