Benchmark Almanac

GDPx

Claude Fable 5.2 16.4% leads since November 2026
Not in model cards

Agents work long tasks in simulated workplaces across 11 domains, including chip design, cybersecurity, biology, finance, quant trading and recursive self-improvement. Fourteen specialist partners built the tasks, with a default limit of 8 hours each. Score is pass@1 over three attempts, with every domain weighted equally.

Version 1.0. Each partner published one example task, and the rest are private. Safety classifiers lower the scores of Opus 5.5 and GPT-6 Astra on cyber, biology and alignment tasks. Finance and legal report partial credit, since every model scores 0 under binary grading. SemiAnalysis plans a new version every 3 to 6 months.

Made by
SemiAnalysis
Released
November 2026
Kind
Tasks from real jobs
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last model added
November 2026
Scores refreshed
November 12th, 2026
21
Entries on its own leaderboard, 20 models we track

Claude Fable 5.2 leads with 16.4%

Best score per entry 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

21 entries. Scroll sideways for the rest.

  1. 16.4 Claude Fable 5.2
  2. 12.0 Claude Opus 5.5
  3. 10.9 Gemini 4 Carbon
  4. 9.60 GPT-6 Astra
  5. 9.00 Muse Wonder Full
  6. 8.53 Claude Fable 5.1
  7. 8.48 GPT-6.1 Terra
  8. 8.43 GPT-6.1 Sol
  9. 8.22 Claude Sonnet 5.5
  10. 7.34 Gemini 4 Argon
  11. 6.82 Muse Wonder
  12. 5.39 DeepSeek 4.1 Flash 1017
  13. 5.11 GPT-6 Luna
  14. 4.94 Claude Haiku 5.5
  15. 4.01 MiniMax-M3.1-Flash
  16. 3.87 Grok 4.7
  17. 3.70 Kimi K3
  18. 3.47 Qwen 3.8 Max
  19. 2.69 GLM-5.3
  20. 2.34 Inkling
  21. 1.67 Muse Spark Open
  1. 1 Claude Fable 5.2 Max 16.4%
  2. 2 Claude Opus 5.5 Max 12%
  3. 3 Gemini 4 Carbon High 10.9%
  4. 4 GPT-6 Astra Max 9.60%
  5. 5 Muse Wonder Full Highest 9.00%
  6. 6 Claude Fable 5.1 Max 8.53%
  7. 7 GPT-6.1 Terra Max 8.48%
  8. 8 GPT-6.1 Sol Max 8.43%
  9. 9 Claude Sonnet 5.5 Max 8.22%
  10. 10 Gemini 4 Argon High 7.34%
  11. 11 Muse Wonder Highest 6.82%
  12. 12 DeepSeek 4.1 Flash 1017 Highest 5.39%
  13. 13 GPT-6 Luna Max 5.11%
  14. 14 Claude Haiku 5.5 Max 4.94%
  15. 15 MiniMax-M3.1-Flash Highest 4.01%
  16. 16 Grok 4.7 xHigh 3.87%
  17. 17 Kimi K3 Highest 3.70%
  18. 18 Qwen 3.8 Max Highest 3.47%
  19. 19 GLM-5.3 Max 2.69%
  20. 20 Inkling Highest 2.34%
  21. 21 Muse Spark Open Highest 1.67%

Source: Official GDPx leaderboard

#EntryScoreSettingDateSource
1 Claude Fable 5.2 16.4% Max, Claude Code Nov 2026 Official leaderboard
2 Claude Opus 5.5 12% Max, Claude Code Sep 2026 Official leaderboard
3 Gemini 4 Carbon 10.9% High, Antigravity Oct 2026 Official leaderboard
4 GPT-6 Astra 9.60% Max, Codex Sep 2026 Official leaderboard
5 Muse Wonder Full 9.00% Highest, Muse Code Oct 2026 Official leaderboard
6 Claude Fable 5.1 8.53% Max, Claude Code Sep 2026 Official leaderboard
7 GPT-6.1 Terra 8.48% Max, Codex Oct 2026 Official leaderboard
8 GPT-6.1 Sol 8.43% Max, Codex Sep 2026 Official leaderboard
9 Claude Sonnet 5.5 8.22% Max, Claude Code Sep 2026 Official leaderboard
10 Gemini 4 Argon 7.34% High, Antigravity Sep 2026 Official leaderboard
11 Muse Wonder 6.82% Highest, Muse Code Oct 2026 Official leaderboard
12 DeepSeek 4.1 Flash 1017 5.39% Highest, DeepSeek Harness Official leaderboard
13 GPT-6 Luna 5.11% Max, Codex Sep 2026 Official leaderboard
14 Claude Haiku 5.5 4.94% Max, Claude Code Oct 2026 Official leaderboard
15 MiniMax-M3.1-Flash 4.01% Highest, MCode Oct 2026 Official leaderboard
16 Grok 4.7 3.87% xHigh, Grok Build Sep 2026 Official leaderboard
17 Kimi K3 3.70% Highest, Kimi Code Jul 2026 Official leaderboard
18 Qwen 3.8 Max 3.47% Highest, OpenCode Aug 2026 Official leaderboard
19 GLM-5.3 2.69% Max, ZCode Aug 2026 Official leaderboard
20 Inkling 2.34% Highest, OpenCode Jul 2026 Official leaderboard
21 Muse Spark Open 1.67% Highest, Muse Code Oct 2026 Official leaderboard

Cost vs. Performance Pareto Frontier

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% $0.1$1$10$100 Average cost per task ($), log scale, lower is better GDPx score Inkling (Highest): GDPx score 2.34%, average cost per task $6.7 Kimi K3 (Highest): GDPx score 3.7%, average cost per task $28.4 Qwen 3.8 Max (Highest): GDPx score 3.47%, average cost per task $11.5 GLM-5.3 (Max): GDPx score 2.69%, average cost per task $11.8 Claude Fable 5.1 (Max): GDPx score 8.53%, average cost per task $67.2 GPT-6 Astra (Max): GDPx score 9.6%, average cost per task $54 Grok 4.7 (xHigh): GDPx score 3.87%, average cost per task $16.3 Claude Sonnet 5.5 (Max): GDPx score 8.22%, average cost per task $45.3 Gemini 4 Argon (High): GDPx score 7.34%, average cost per task $12.8 Claude Haiku 5.5 (Max): GDPx score 4.94%, average cost per task $3.98 Muse Wonder (Highest): GDPx score 6.82%, average cost per task $15.6 Muse Spark Open (Highest): GDPx score 1.67%, average cost per task $4.35 MiniMax-M3.1-Flash (Highest): GDPx score 4.01%, average cost per task $1.61 Claude Opus 5.5 (Max): GDPx score 12%, average cost per task $51.4 GPT-6 Luna (Max): GDPx score 5.11%, average cost per task $0.92 GPT-6.1 Sol (Max): GDPx score 8.43%, average cost per task $9.8 GPT-6.1 Terra (Max): GDPx score 8.48%, average cost per task $15 Gemini 4 Carbon (High): GDPx score 10.9%, average cost per task $16.3 Claude Fable 5.2 (Max): GDPx score 16.4%, average cost per task $88.2 DeepSeek 4.1 Flash 1017 (Highest): GDPx score 5.39%, average cost per task $0.94. Not in our model list Claude Fable 5.2 (Max)Claude Opus 5.5 (Max)Gemini 4 Carbon (High)GPT-6.1 Terra (Max)GPT-6.1 Sol (Max)DeepSeek 4.1 Flash 1017 (Highest)GPT-6 Luna (Max) 0% 5% 10% 15% 20% $0.1$1$10$100 Average cost per task ($), log scale, lower is better GDPx score Inkling (Highest): GDPx score 2.34%, average cost per task $6.7 Kimi K3 (Highest): GDPx score 3.7%, average cost per task $28.4 Qwen 3.8 Max (Highest): GDPx score 3.47%, average cost per task $11.5 GLM-5.3 (Max): GDPx score 2.69%, average cost per task $11.8 Claude Fable 5.1 (Max): GDPx score 8.53%, average cost per task $67.2 GPT-6 Astra (Max): GDPx score 9.6%, average cost per task $54 Grok 4.7 (xHigh): GDPx score 3.87%, average cost per task $16.3 Claude Sonnet 5.5 (Max): GDPx score 8.22%, average cost per task $45.3 Gemini 4 Argon (High): GDPx score 7.34%, average cost per task $12.8 Claude Haiku 5.5 (Max): GDPx score 4.94%, average cost per task $3.98 Muse Wonder (Highest): GDPx score 6.82%, average cost per task $15.6 Muse Spark Open (Highest): GDPx score 1.67%, average cost per task $4.35 MiniMax-M3.1-Flash (Highest): GDPx score 4.01%, average cost per task $1.61 Claude Opus 5.5 (Max): GDPx score 12%, average cost per task $51.4 GPT-6 Luna (Max): GDPx score 5.11%, average cost per task $0.92 GPT-6.1 Sol (Max): GDPx score 8.43%, average cost per task $9.8 GPT-6.1 Terra (Max): GDPx score 8.48%, average cost per task $15 Gemini 4 Carbon (High): GDPx score 10.9%, average cost per task $16.3 Claude Fable 5.2 (Max): GDPx score 16.4%, average cost per task $88.2 DeepSeek 4.1 Flash 1017 (Highest): GDPx score 5.39%, average cost per task $0.94. Not in our model list Claude Fable 5.2 (Max)Claude Opus 5.5 (Max)Gemini 4 Carbon (High)GPT-6.1 Terra (Max)GPT-6.1 Sol (Max)DeepSeek 4.1 Flash 1017 (Highest)GPT-6 Luna (Max)
ModelSettingGDPx scoreAverage cost per task
Claude Fable 5.2 Max, Claude Code 16.4% $88.15
Claude Opus 5.5 Max, Claude Code 12% $51.41
Gemini 4 Carbon High, Antigravity 10.9% $16.29
GPT-6 Astra Max, Codex 9.6% $53.99
Claude Fable 5.1 Max, Claude Code 8.5% $67.16
GPT-6.1 Terra Max, Codex 8.5% $15.02
GPT-6.1 Sol Max, Codex 8.4% $9.8
Claude Sonnet 5.5 Max, Claude Code 8.2% $45.27
Gemini 4 Argon High, Antigravity 7.3% $12.79
Muse Wonder Highest, Muse Code 6.8% $15.63
DeepSeek 4.1 Flash 1017 Highest, DeepSeek Harness 5.4% $0.94
GPT-6 Luna Max, Codex 5.1% $0.92
Claude Haiku 5.5 Max, Claude Code 4.9% $3.98
MiniMax-M3.1-Flash Highest, MCode 4% $1.61
Grok 4.7 xHigh, Grok Build 3.9% $16.34
Kimi K3 Highest, Kimi Code 3.7% $28.38
Qwen 3.8 Max Highest, OpenCode 3.5% $11.53
GLM-5.3 Max, ZCode 2.7% $11.79
Inkling Highest, OpenCode 2.3% $6.7
Muse Spark Open Highest, Muse Code 1.7% $4.35

The best score, 16.4%, costs $88 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.

Anthropic OpenAI Google DeepMind Cost frontier One model at each effort level
0% 5% 10% 15% 20% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.2, Max Claude Opus 5.5, Max Gemini 4 Carbon, High GPT-6.1 Terra, Max GPT-6 Luna, Max 0% 5% 10% 15% 20% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.2, Max Claude Opus 5.5, Max

Source: Official GDPx leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.2 Max Claude Code 16.4% $88 1,226,025
Claude Opus 5.5 Max Claude Code 12% $51 548,088
Gemini 4 Carbon High Antigravity 10.9% $16 594,284
GPT-6 Astra Max Codex 9.60% $54 275,129
Muse Wonder Full Highest Muse Code 9.00% 212,053
Claude Fable 5.1 Max Claude Code 8.53% $67 418,002
GPT-6.1 Terra Max Codex 8.48% $15 548,048
GPT-6.1 Sol Max Codex 8.43% $9.80 351,974
Claude Sonnet 5.5 Max Claude Code 8.22% $45 967,119
Gemini 4 Argon High Antigravity 7.34% $13 358,521
Muse Wonder Highest Muse Code 6.82% $16 215,188
GPT-6 Luna Max Codex 5.11% $0.92 438,311
Claude Haiku 5.5 Max Claude Code 4.94% $3.98 710,184
MiniMax-M3.1-Flash Highest MCode 4.01% $1.61 296,857
Grok 4.7 xHigh Grok Build 3.87% $16 145,436
Kimi K3 Highest Kimi Code 3.70% $28 254,486
Qwen 3.8 Max Highest OpenCode 3.47% $12 363,420
GLM-5.3 Max ZCode 2.69% $12 211,569
Inkling Highest OpenCode 2.34% $6.70 59,848
Muse Spark Open Highest Muse Code 1.67% $4.35 199,579

2 models have held number one on the official leaderboard since its release in November 2026

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

Anthropic
Own leaderboard 2 leaders
Claude Fable 5.2 16.4%

Source: Official GDPx 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 Fable 5.2 Claude Opus 5.5 Kimi K3 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Claude Fable 5.2 GPT-6 Astra

Source: Official GDPx 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%