Benchmark Almanac

Vals Multimodal Index

Claude Fable 5 74.2% leads since June 2026
Not in model cards

Models are scored on a GDP-weighted mix of seven Vals finance, coding and education benchmarks, including MortgageTax and SAGE, which grades handwritten math work.

Finance (weight 2.0) averages CorpFin, Finance Agent v2 and MortgageTax. Coding (1.4) mixes SWE-bench Verified, Terminal-Bench 2.1 and Vibe Code Bench. Introduced October 29th, 2025, version 1.2 by August 11th, 2026.

Made by
Vals AI
Released
October 2025
Kind
Composite indices
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Opus 5 #10 on the capability index
Last model added
August 2026
Scores refreshed
November 12th, 2026
33
Models scored on its own leaderboard

Claude Fable 5 leads with 74.2% among models released by August 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 August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

33 models. Scroll sideways for the rest.

  1. 74.2 Claude Fable 5
  2. 73.9 Claude Opus 5
  3. 73.4 Kimi K3
  4. 72.6 GPT-5.6 Sol
  5. 70.9 Claude Opus 4.8
  6. 69.8 Muse Spark 1.2
  7. 69.1 GPT-5.6 Luna
  8. 68.8 Claude Sonnet 5
  9. 68.1 GPT-5.5
  10. 67.4 Claude Opus 4.7
  11. 66.7 Muse Spark 1.1
  12. 65.4 Qwen 3.8 Max
  13. 65.1 Gemini 3.6 Flash
  14. 65.1 GPT-5.6 Terra
  15. 63.4 Grok 4.5
  16. 62.9 Gemini 3.5 Flash
  17. 60.6 Claude Sonnet 4.6
  18. 60.0 MiniMax-M3
  19. 56.4 Kimi K2.6
  20. 56.1 Gemini 3.1 Pro
  21. 54.2 GPT-5.4 Mini
  22. 53.9 Qwen3.7-Plus
  23. 52.8 MiMo-V2.5
  24. 52.2 Gemini 3 Flash
  25. 51.5 Qwen 3.6 Plus
  26. 50.0 Inkling-Small
  27. 49.4 Inkling
  28. 47.6 GPT-5.4 Nano
  29. 43.3 Grok 4.3
  30. 42.9 Claude Haiku 4.5
  31. 41.4 Gemini 3.1 Flash-Lite
  32. 39.1 Grok 4.20
  33. 34.8 Mistral Medium 3.5
  1. 1 Claude Fable 5 74.2%
  2. 2 Claude Opus 5 73.9%
  3. 3 Kimi K3 73.4%
  4. 4 GPT-5.6 Sol 72.6%
  5. 5 Claude Opus 4.8 70.9%
  6. 6 Muse Spark 1.2 69.8%
  7. 7 GPT-5.6 Luna 69.1%
  8. 8 Claude Sonnet 5 68.8%
  9. 9 GPT-5.5 68.1%
  10. 10 Claude Opus 4.7 67.4%
  11. 11 Muse Spark 1.1 66.7%
  12. 12 Qwen 3.8 Max 65.4%
  13. 13 Gemini 3.6 Flash 65.1%
  14. 14 GPT-5.6 Terra 65.1%
  15. 15 Grok 4.5 63.4%
  16. 16 Gemini 3.5 Flash 62.9%
  17. 17 Claude Sonnet 4.6 60.6%
  18. 18 MiniMax-M3 60%
  19. 19 Kimi K2.6 56.4%
  20. 20 Gemini 3.1 Pro 56.1%
  21. 21 GPT-5.4 Mini 54.2%
  22. 22 Qwen3.7-Plus 53.9%
  23. 23 MiMo-V2.5 52.8%
  24. 24 Gemini 3 Flash 52.2%
  25. 25 Qwen 3.6 Plus 51.5%
  26. 26 Inkling-Small 50%
  27. 27 Inkling 49.4%
  28. 28 GPT-5.4 Nano 47.6%
  29. 29 Grok 4.3 43.3%
  30. 30 Claude Haiku 4.5 thinking 42.9%
  31. 31 Gemini 3.1 Flash-Lite 41.4%
  32. 32 Grok 4.20 reasoning 39.1%
  33. 33 Mistral Medium 3.5 34.8%

Source: Official Vals Multimodal Index leaderboard

#ModelScoreSettingDateSource
1 Claude Fable 5 74.2% Jun 2026 Official leaderboard
2 Claude Opus 5 73.9% Jul 2026 Official leaderboard
3 Kimi K3 73.4% Jul 2026 Official leaderboard
4 GPT-5.6 Sol 72.6% Jul 2026 Official leaderboard
5 Claude Opus 4.8 70.9% May 2026 Official leaderboard
6 Muse Spark 1.2 69.8% Aug 2026 Official leaderboard
7 GPT-5.6 Luna 69.1% Jul 2026 Official leaderboard
8 Claude Sonnet 5 68.8% Jun 2026 Official leaderboard
9 GPT-5.5 68.1% Apr 2026 Official leaderboard
10 Claude Opus 4.7 67.4% Apr 2026 Official leaderboard
11 Muse Spark 1.1 66.7% Jul 2026 Official leaderboard
12 Qwen 3.8 Max 65.4% Aug 2026 Official leaderboard
13 Gemini 3.6 Flash 65.1% Jul 2026 Official leaderboard
14 GPT-5.6 Terra 65.1% Jul 2026 Official leaderboard
15 Grok 4.5 63.4% Jul 2026 Official leaderboard
16 Gemini 3.5 Flash 62.9% May 2026 Official leaderboard
17 Claude Sonnet 4.6 60.6% Feb 2026 Official leaderboard
18 MiniMax-M3 60% Jun 2026 Official leaderboard
19 Kimi K2.6 56.4% Apr 2026 Official leaderboard
20 Gemini 3.1 Pro 56.1% Feb 2026 Official leaderboard
21 GPT-5.4 Mini 54.2% Mar 2026 Official leaderboard
22 Qwen3.7-Plus 53.9% Jun 2026 Official leaderboard
23 MiMo-V2.5 52.8% Apr 2026 Official leaderboard
24 Gemini 3 Flash 52.2% Dec 2025 Official leaderboard
25 Qwen 3.6 Plus 51.5% Mar 2026 Official leaderboard
26 Inkling-Small 50% Jul 2026 Official leaderboard
27 Inkling 49.4% Jul 2026 Official leaderboard
28 GPT-5.4 Nano 47.6% Mar 2026 Official leaderboard
29 Grok 4.3 43.3% Apr 2026 Official leaderboard
30 Claude Haiku 4.5 42.9% thinking Oct 2025 Official leaderboard
31 Gemini 3.1 Flash-Lite 41.4% Mar 2026 Official leaderboard
32 Grok 4.20 39.1% reasoning Feb 2026 Official leaderboard
33 Mistral Medium 3.5 34.8% Apr 2026 Official leaderboard

Accuracy against latency

The Vals scatter chart with its x axis switched from Cost to Latency: mean time per test against accuracy, on a log scale. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

40% 60% 80% 1001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Claude Haiku 4.5 (thinking): Accuracy 42.9%, latency (mean per test) 283 s Gemini 3 Flash: Accuracy 52.2%, latency (mean per test) 260 s Claude Sonnet 4.6: Accuracy 60.6%, latency (mean per test) 536 s Grok 4.20 (reasoning): Accuracy 39.1%, latency (mean per test) 171 s Gemini 3.1 Pro: Accuracy 56.1%, latency (mean per test) 323 s GPT-5.4 Mini: Accuracy 54.2%, latency (mean per test) 788 s GPT-5.4 Nano: Accuracy 47.6%, latency (mean per test) 671 s Qwen 3.6 Plus: Accuracy 51.5%, latency (mean per test) 643 s Claude Opus 4.7: Accuracy 67.4%, latency (mean per test) 564 s Kimi K2.6: Accuracy 56.4%, latency (mean per test) 904 s GPT-5.5: Accuracy 68.1%, latency (mean per test) 504 s MiMo-V2.5: Accuracy 52.8%, latency (mean per test) 430 s Mistral Medium 3.5: Accuracy 34.8%, latency (mean per test) 735 s Grok 4.3: Accuracy 43.3%, latency (mean per test) 339 s Gemini 3.5 Flash: Accuracy 62.9%, latency (mean per test) 266 s Claude Opus 4.8: Accuracy 70.9%, latency (mean per test) 980 s MiniMax-M3: Accuracy 60%, latency (mean per test) 1,076 s Qwen3.7-Plus: Accuracy 53.9%, latency (mean per test) 629 s Claude Sonnet 5: Accuracy 68.8%, latency (mean per test) 1,012 s Grok 4.5: Accuracy 63.4%, latency (mean per test) 319 s Inkling: Accuracy 49.4%, latency (mean per test) 929 s Inkling-Small: Accuracy 50%, latency (mean per test) 709 s Kimi K3: Accuracy 73.4%, latency (mean per test) 906 s Gemini 3.6 Flash: Accuracy 65.1%, latency (mean per test) 536 s Claude Opus 5: Accuracy 73.9%, latency (mean per test) 854 s Qwen 3.8 Max: Accuracy 65.4%, latency (mean per test) 1,996 s Gemini 3.1 Flash-Lite: Accuracy 41.4%, latency (mean per test) 152 s Claude Fable 5: Accuracy 74.2%, latency (mean per test) 769 s GPT-5.6 Luna: Accuracy 69.1%, latency (mean per test) 407 s GPT-5.6 Sol: Accuracy 72.6%, latency (mean per test) 551 s GPT-5.6 Terra: Accuracy 65.1%, latency (mean per test) 258 s Muse Spark 1.1: Accuracy 66.7%, latency (mean per test) 288 s Muse Spark 1.2: Accuracy 69.8%, latency (mean per test) 446 s Claude Fable 5GPT-5.6 SolMuse Spark 1.2GPT-5.6 LunaMuse Spark 1.1GPT-5.6 TerraGemini 3.1 Flash-Lite 40% 60% 80% 1001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Claude Haiku 4.5 (thinking): Accuracy 42.9%, latency (mean per test) 283 s Gemini 3 Flash: Accuracy 52.2%, latency (mean per test) 260 s Claude Sonnet 4.6: Accuracy 60.6%, latency (mean per test) 536 s Grok 4.20 (reasoning): Accuracy 39.1%, latency (mean per test) 171 s Gemini 3.1 Pro: Accuracy 56.1%, latency (mean per test) 323 s GPT-5.4 Mini: Accuracy 54.2%, latency (mean per test) 788 s GPT-5.4 Nano: Accuracy 47.6%, latency (mean per test) 671 s Qwen 3.6 Plus: Accuracy 51.5%, latency (mean per test) 643 s Claude Opus 4.7: Accuracy 67.4%, latency (mean per test) 564 s Kimi K2.6: Accuracy 56.4%, latency (mean per test) 904 s GPT-5.5: Accuracy 68.1%, latency (mean per test) 504 s MiMo-V2.5: Accuracy 52.8%, latency (mean per test) 430 s Mistral Medium 3.5: Accuracy 34.8%, latency (mean per test) 735 s Grok 4.3: Accuracy 43.3%, latency (mean per test) 339 s Gemini 3.5 Flash: Accuracy 62.9%, latency (mean per test) 266 s Claude Opus 4.8: Accuracy 70.9%, latency (mean per test) 980 s MiniMax-M3: Accuracy 60%, latency (mean per test) 1,076 s Qwen3.7-Plus: Accuracy 53.9%, latency (mean per test) 629 s Claude Sonnet 5: Accuracy 68.8%, latency (mean per test) 1,012 s Grok 4.5: Accuracy 63.4%, latency (mean per test) 319 s Inkling: Accuracy 49.4%, latency (mean per test) 929 s Inkling-Small: Accuracy 50%, latency (mean per test) 709 s Kimi K3: Accuracy 73.4%, latency (mean per test) 906 s Gemini 3.6 Flash: Accuracy 65.1%, latency (mean per test) 536 s Claude Opus 5: Accuracy 73.9%, latency (mean per test) 854 s Qwen 3.8 Max: Accuracy 65.4%, latency (mean per test) 1,996 s Gemini 3.1 Flash-Lite: Accuracy 41.4%, latency (mean per test) 152 s Claude Fable 5: Accuracy 74.2%, latency (mean per test) 769 s GPT-5.6 Luna: Accuracy 69.1%, latency (mean per test) 407 s GPT-5.6 Sol: Accuracy 72.6%, latency (mean per test) 551 s GPT-5.6 Terra: Accuracy 65.1%, latency (mean per test) 258 s Muse Spark 1.1: Accuracy 66.7%, latency (mean per test) 288 s Muse Spark 1.2: Accuracy 69.8%, latency (mean per test) 446 s Claude Fable 5GPT-5.6 SolMuse Spark 1.2GPT-5.6 LunaMuse Spark 1.1GPT-5.6 TerraGemini 3.1 Flash-Lite
ModelSettingAccuracyLatency (mean per test)
Claude Fable 5 74.2% 769.49 s
Claude Opus 5 73.9% 853.76 s
Kimi K3 73.4% 906.43 s
GPT-5.6 Sol 72.6% 550.77 s
Claude Opus 4.8 70.9% 979.85 s
Muse Spark 1.2 69.8% 445.98 s
GPT-5.6 Luna 69.1% 406.66 s
Claude Sonnet 5 68.8% 1,012 s
GPT-5.5 68.1% 503.75 s
Claude Opus 4.7 67.4% 564.05 s
Muse Spark 1.1 66.7% 288.19 s
Qwen 3.8 Max 65.4% 1,996 s
Gemini 3.6 Flash 65.1% 535.65 s
GPT-5.6 Terra 65.1% 258.09 s
Grok 4.5 63.4% 318.5 s
Gemini 3.5 Flash 62.9% 266.49 s
Claude Sonnet 4.6 60.6% 536.04 s
MiniMax-M3 60% 1,076 s
Kimi K2.6 56.4% 903.81 s
Gemini 3.1 Pro 56.1% 322.91 s
GPT-5.4 Mini 54.2% 788.34 s
Qwen3.7-Plus 53.9% 628.66 s
MiMo-V2.5 52.8% 430.31 s
Gemini 3 Flash 52.2% 260.28 s
Qwen 3.6 Plus 51.5% 642.88 s
Inkling-Small 50.1% 709.05 s
Inkling 49.4% 929.08 s
GPT-5.4 Nano 47.6% 671.05 s
Grok 4.3 43.3% 339.04 s
Claude Haiku 4.5 thinking 42.9% 282.99 s
Gemini 3.1 Flash-Lite 41.4% 152.43 s
Grok 4.20 reasoning 39.1% 170.99 s
Mistral Medium 3.5 34.8% 734.7 s

The best score, 74.2%, costs $7.97 a task

Runs on the cost frontier score higher than any run that's cheaper. Muse Spark 1.2 comes within 4.4 points of the best for $0.51 a task, 16 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Xiaomi Corp Anthropic OpenAI Moonshot AI Meta Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5 Kimi K3 Muse Spark 1.2 GPT-5.6 Luna MiMo-V2.5 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5 Muse Spark 1.2

Source: Official Vals Multimodal Index leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5 74.2% $7.97
Claude Opus 5 73.9% $6.14
Kimi K3 73.4% $1.70
GPT-5.6 Sol 72.6% $5.36
Claude Opus 4.8 70.9% $5.46
Muse Spark 1.2 69.8% $0.51
GPT-5.6 Luna 69.1% $0.16
Claude Sonnet 5 68.8% $6.53
GPT-5.5 68.1% $3.32
Claude Opus 4.7 67.4% $4.44
Muse Spark 1.1 66.7% $0.36
Qwen 3.8 Max 65.4% $1.93
Gemini 3.6 Flash 65.1% $1.02
GPT-5.6 Terra 65.1% $0.80
Grok 4.5 63.4% $0.91
Gemini 3.5 Flash 62.9% $1.01
Claude Sonnet 4.6 60.6% $1.58
MiniMax-M3 60% $1.08
Kimi K2.6 56.4% $0.52
Gemini 3.1 Pro 56.1% $0.98
GPT-5.4 Mini 54.2% $0.49
Qwen3.7-Plus 53.9% $0.26
MiMo-V2.5 52.8% $0.03
Gemini 3 Flash 52.2% $0.29
Qwen 3.6 Plus 51.5% $1.12
Inkling-Small 50% $0.14
Inkling 49.4% $0.63
GPT-5.4 Nano 47.6% $0.23
Grok 4.3 43.3% $0.50
Claude Haiku 4.5 thinking 42.9% $0.41
Gemini 3.1 Flash-Lite 41.4% $0.12
Grok 4.20 reasoning 39.1% $0.40
Mistral Medium 3.5 34.8% $3.45

7 models have held number one on the official leaderboard since its release in October 2025

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

Anthropic Google DeepMind OpenAI
Own leaderboard Not tested since Aug 2026
Claude Fable 5 74.2%

Source: Official Vals Multimodal Index 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 GPT-5.5 Claude Sonnet 4.6 Gemini 3 Flash Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5 Claude Opus 4.7

Source: Official Vals Multimodal Index 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 Other

All 43 in Other
  • Personal Superintelligence Index (Meta) In use Meta's composite of seven personal and knowledge-work evals: Agents' Last Exam, AutomationBench-AA, FocusBench, JobBench, GDP.pdf, VBT and DRACO.
  • Artificial Analysis Intelligence Index v5 In use Version 5 of Artificial Analysis's index: nine evaluations in four equal categories, most with private test sets, run three times each in its own agent harness. Gemini 4 Carbon 59.0
  • SAGE v1.1 (Vals) In use Models grade handwritten undergraduate math and computer science work against rubrics. Scored by class-balanced accuracy of each rubric check against expert grading. Claude Opus 5.5 91.3%
  • MTEB Fading Massive Text Embedding Benchmark: retrieval, clustering, classification across languages and modalities. sonar 42%
  • Anthropic ECI (AECI) Only in system cards Index that combines a basket of capability benchmarks into one capability score to track trends across Anthropic models. Claude Opus 5.5 169.4
  • SPEED-Bench Only in system cards Measures speculative decoding across varied prompts and domains. Reports the average acceptance length of draft tokens, so higher means faster generation at the same output. Nemotron 3 Ultra 5.15