Benchmark Almanac

Vals Finance Agent v2.1

Claude Fable 5.2 69.5% leads since November 2026
Claude Fable 5.2 69.5% leads since November 2026
In use
Version v2.1

The official leaderboard hasn’t scored a model released after Gemini 4 Neon (November 2026). Labs report their own numbers for one newer model: Inkling 2 Mini at 56.5%.

Agents answer analyst-grade questions on public company filings, scored with dealbreaker-gated partial credit. A three-model jury grades answers and fails hedged ones.

Revises v2 with question and rubric fixes and a jury of GPT-6.1 Sol, Gemini 3.8 Flash and Claude Sonnet 5.5. Vals says scores are not comparable with v2. Part of Vals Index v2.2 from October 19th, 2026.

Made by
Vals AI
Released
October 2026
Kind
Finance
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text, Documents → Actions, Text
Interaction
Agentic
Environment
Browser
Run by
A host
Occupations
Business and financial operations
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 19th, 2026

In the authors’ words

Artificial Intelligence (AI) technology has emerged as a transformative force in financial analysis and the finance industry, though significant questions remain about the full capabilities of Large Language Model (LLM) agents in this domain. We present the Finance Agent Benchmark, featuring challenging and diverse real-world finance research problems that require LLMs to perform complex analysis using recent SEC filings. We construct the benchmark using a taxonomy of nine financial task categories, developed in consultation with experts from banks, hedge funds, and private equity firms. The dataset includes 537 expert-authored questions covering tasks from information retrieval to complex financial modeling, each validated through a rigorous review process to ensure accuracy and relevance. Moreover, we implement an agentic harness that equips LLMs with tools sufficient to produce accurate responses, including Google Search and EDGAR database access. Overall, the Finance Agent Benchmark provides a comprehensive testbed for measuring the progress of LLM-driven finance agents. Our evaluation reveals significant limitations in current AI capabilities - even the best-performing model (OpenAI o3) achieved only 46.8% accuracy at an average cost of $3.79 per query. This underscores the need for further advancements before reliable deployment in high-stakes finance settings.

Abstract of Finance Agent Benchmark: Benchmarking LLMs on Real-world Financial Research Tasks (arXiv:2508.00828, CC0)
45
Entries on its own leaderboard, 44 models we track
1
Launch post lists it, from 1 lab
1
System card reports it, from 1 lab

Claude Fable 5.2 leads with 69.5% Claude Fable 5.2 leads with 69.5%

Best score per entry from the benchmark's own leaderboard, read on November 19th, 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. This view shows only models we track, each at its best.

Claude Fable 5.2 leads with 69.5%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.

  1. 1 Claude Fable 5.2 69.5%
  2. 2 Gemini 4 Neon 68.6%
  3. 3 Claude Opus 5.5 67.8%
  4. 4 Gemini 4 Argon 67.7%
  5. 5 Muse Wonder Full 67.6%
  6. 6 Inkling 2 Small 67.5%
  7. 7 Claude Sonnet 5.5 67.3%
  8. 8 Gemini 4 Carbon 67.2%
  9. 9 Qwen4-Max 66.7%
  10. 10 Gemini 3.8 Flash 65.8%
  11. 11 Muse Wonder 65.5%
  12. 12 DeepSeek V4.1 Flash 65.4%
  13. 13 Claude Haiku 5.5 65.4%
  14. 14 Gemini 4 Hydrogen 65.1%
  15. 15 GLM-5.3 65%
  16. 16 Tencent Hy4 64.7%
  17. 17 MiMo-V2.6 Pro 64.3%
  18. 18 Muse Ember 63.8%
  19. 19 MiMo-V2.6 Flash 63.3%
  20. 20 DeepSeek V4.1 Flash 1017 62.9%
  21. 21 GPT-6.1 Sol 61.8%
  22. 22 Mistral Large 4 61.7%
  23. 23 Tencent Hy4 preview 61.6%
  24. 24 Kimi K3.1 Pro 61.5%
  25. 25 Mistral Large 4 Preview 61.4%
  26. 26 Hy4 Flash 60.8%
  27. 27 GPT-6.2 Astra 60.8%
  28. 28 Muse Spark Open 60.7%
  29. 29 GLM-5.3-Flash 60.6%
  30. 30 Grok 4.7 60.6%
  31. 31 GPT-6 Astra 60.4%
  32. 32 Kimi K3 60%
  33. 33 GPT-6.2 Sol 58.9%
  34. 34 Step 5 Preview 58.9%
  35. 35 Qwen3.8 Max (0902) 58.8%
  36. 36 MiniMax-M3.1-Flash 58.8%
  37. 37 DeepSeek V4 Pro 0813 57.1%
  38. 38 GPT-6 Luna 56.8%
  39. 39 GPT-6.1 Terra 56%
  40. 40 Mercury 3 54.7%
  41. 41 MiniMax-M3 54.3%
  42. 42 Beam 1.0 51.8%
  43. 43 Inkling 49.9%
  44. 44 Nemotron 3.5 Super VL 47.7%
  45. 45 Inkling-Small 44.8%

Source: Official Vals Finance Agent v2.1 leaderboard

Source: Official Vals Finance Agent v2.1 leaderboard

Sources: Official Vals Finance Agent v2.1 leaderboard and labs’ model cards

#EntryScore ± 95% CISettingDateSource
1 Claude Fable 5.2 69.5% ±3.9 Nov 2026 Official leaderboard
2 Gemini 4 Neon 68.6% ±3.8 Nov 2026 Official leaderboard
3 Claude Opus 5.5 67.8% ±4 Sep 2026 Official leaderboard
4 Gemini 4 Argon 67.7% ±3.9 Sep 2026 Official leaderboard
5 Muse Wonder Full 67.6% ±3.9 Oct 2026 Official leaderboard
6 Inkling 2 Small 67.5% ±4 Nov 2026 Official leaderboard
7 Claude Sonnet 5.5 67.3% ±4 Sep 2026 Official leaderboard
8 Gemini 4 Carbon 67.2% ±4 Oct 2026 Official leaderboard
9 Qwen4-Max 66.7% ±4 Nov 2026 Official leaderboard
10 Gemini 3.8 Flash 65.8% ±4 Sep 2026 Official leaderboard
11 Muse Wonder 65.5% ±4 Oct 2026 Official leaderboard
12 DeepSeek V4.1 Flash 65.4% ±4 Sep 2026 Official leaderboard
13 Claude Haiku 5.5 65.4% ±4 Oct 2026 Official leaderboard
14 Gemini 4 Hydrogen 65.1% ±4 Nov 2026 Official leaderboard
15 GLM-5.3 65% ±3.9 Aug 2026 Official leaderboard
16 Tencent Hy4 64.7% ±4 Nov 2026 Official leaderboard
17 MiMo-V2.6 Pro 64.3% ±3.9 Sep 2026 Official leaderboard
18 Muse Ember 63.8% ±3.9 Nov 2026 Official leaderboard
19 MiMo-V2.6 Flash 63.3% ±4.1 Sep 2026 Official leaderboard
20 DeepSeek V4.1 Flash 1017 62.9% ±4 Oct 2026 Official leaderboard
21 GPT-6.1 Sol 61.8% ±4.1 Sep 2026 Official leaderboard
22 Mistral Large 4 61.7% ±4 Oct 2026 Official leaderboard
23 Tencent Hy4 preview 61.6% ±4 Aug 2026 Official leaderboard
24 Kimi K3.1 Pro 61.5% ±4.1 Oct 2026 Official leaderboard
25 Mistral Large 4 Preview 61.4% ±4 Oct 2026 Official leaderboard
26 Hy4 Flash 60.8% ±4.1 Official leaderboard
27 GPT-6.2 Astra 60.8% ±4.2 Nov 2026 Official leaderboard
28 Muse Spark Open 60.7% ±4 Oct 2026 Official leaderboard
29 GLM-5.3-Flash 60.6% ±4 Aug 2026 Official leaderboard
30 Grok 4.7 60.6% ±4.1 Sep 2026 Official leaderboard
31 GPT-6 Astra 60.4% ±4.2 Sep 2026 Official leaderboard
32 Kimi K3 60% ±4.1 Jul 2026 Official leaderboard
33 GPT-6.2 Sol 58.9% ±4.1 Nov 2026 Official leaderboard
34 Step 5 Preview 58.9% ±4 Sep 2026 Official leaderboard
35 Qwen3.8 Max (0902) 58.8% ±4.1 Sep 2026 Official leaderboard
36 MiniMax-M3.1-Flash 58.8% ±4.1 Oct 2026 Official leaderboard
37 DeepSeek V4 Pro 0813 57.1% ±4.1 Aug 2026 Official leaderboard
38 GPT-6 Luna 56.8% ±4.2 Sep 2026 Official leaderboard
39 GPT-6.1 Terra 56% ±4.1 Oct 2026 Official leaderboard
40 Mercury 3 54.7% ±3.9 Oct 2026 Official leaderboard
41 MiniMax-M3 54.3% ±4 Jun 2026 Official leaderboard
42 Beam 1.0 51.8% ±4.1 Oct 2026 Official leaderboard
43 Inkling 49.9% ±4.1 Jul 2026 Official leaderboard
44 Nemotron 3.5 Super VL 47.7% ±4.1 Oct 2026 Official leaderboard
45 Inkling-Small 44.8% ±4.1 Jul 2026 Official leaderboard
#ModelScore ± 95% CISettingDateSource
1 Claude Fable 5.2 69.5% ±3.9 Nov 2026 Official leaderboard
2 Gemini 4 Neon 68.6% ±3.8 Nov 2026 Official leaderboard
3 Claude Opus 5.5 67.8% ±4 Sep 2026 Official leaderboard
4 Gemini 4 Argon 67.7% ±3.9 Sep 2026 Official leaderboard
5 Muse Wonder Full 67.6% ±3.9 Oct 2026 Official leaderboard
6 Inkling 2 Small 67.5% ±4 Nov 2026 Official leaderboard
7 Claude Sonnet 5.5 67.3% ±4 Sep 2026 Official leaderboard
8 Gemini 4 Carbon 67.2% ±4 Oct 2026 Official leaderboard
9 Qwen4-Max 66.7% ±4 Nov 2026 Official leaderboard
10 Gemini 3.8 Flash 65.8% ±4 Sep 2026 Official leaderboard
11 Muse Wonder 65.5% ±4 Oct 2026 Official leaderboard
12 DeepSeek V4.1 Flash 65.4% ±4 Sep 2026 Official leaderboard
13 Claude Haiku 5.5 65.4% ±4 Oct 2026 Official leaderboard
14 Gemini 4 Hydrogen 65.1% ±4 Nov 2026 Official leaderboard
15 GLM-5.3 65% ±3.9 Aug 2026 Official leaderboard
16 Tencent Hy4 64.7% ±4 Nov 2026 Official leaderboard
17 MiMo-V2.6 Pro 64.3% ±3.9 Sep 2026 Official leaderboard
18 Muse Ember 63.8% ±3.9 Nov 2026 Official leaderboard
19 MiMo-V2.6 Flash 63.3% ±4.1 Sep 2026 Official leaderboard
20 DeepSeek V4.1 Flash 1017 62.9% ±4 Oct 2026 Official leaderboard
21 GPT-6.1 Sol 61.8% ±4.1 Sep 2026 Official leaderboard
22 Mistral Large 4 61.7% ±4 Oct 2026 Official leaderboard
23 Tencent Hy4 preview 61.6% ±4 Aug 2026 Official leaderboard
24 Kimi K3.1 Pro 61.5% ±4.1 Oct 2026 Official leaderboard
25 Mistral Large 4 Preview 61.4% ±4 Oct 2026 Official leaderboard
26 GPT-6.2 Astra 60.8% ±4.2 Nov 2026 Official leaderboard
27 Muse Spark Open 60.7% ±4 Oct 2026 Official leaderboard
28 GLM-5.3-Flash 60.6% ±4 Aug 2026 Official leaderboard
29 Grok 4.7 60.6% ±4.1 Sep 2026 Official leaderboard
30 GPT-6 Astra 60.4% ±4.2 Sep 2026 Official leaderboard
31 Kimi K3 60% ±4.1 Jul 2026 Official leaderboard
32 GPT-6.2 Sol 58.9% ±4.1 Nov 2026 Official leaderboard
33 Step 5 Preview 58.9% ±4 Sep 2026 Official leaderboard
34 Qwen3.8 Max (0902) 58.8% ±4.1 Sep 2026 Official leaderboard
35 MiniMax-M3.1-Flash 58.8% ±4.1 Oct 2026 Official leaderboard
36 DeepSeek V4 Pro 0813 57.1% ±4.1 Aug 2026 Official leaderboard
37 GPT-6 Luna 56.8% ±4.2 Sep 2026 Official leaderboard
38 GPT-6.1 Terra 56% ±4.1 Oct 2026 Official leaderboard
39 Mercury 3 54.7% ±3.9 Oct 2026 Official leaderboard
40 MiniMax-M3 54.3% ±4 Jun 2026 Official leaderboard
41 Beam 1.0 51.8% ±4.1 Oct 2026 Official leaderboard
42 Inkling 49.9% ±4.1 Jul 2026 Official leaderboard
43 Nemotron 3.5 Super VL 47.7% ±4.1 Oct 2026 Official leaderboard
44 Inkling-Small 44.8% ±4.1 Jul 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Fable 5.2 69.5% Nov 2026 Official leaderboard
2 Gemini 4 Neon 68.6% Nov 2026 Official leaderboard
3 Claude Opus 5.5 67.8% Sep 2026 Official leaderboard
4 Gemini 4 Argon 67.7% Sep 2026 Official leaderboard
5 Muse Wonder Full 67.6% Oct 2026 Official leaderboard
6 Inkling 2 Small 67.5% Nov 2026 Official leaderboard
7 Claude Sonnet 5.5 67.3% Sep 2026 Official leaderboard
8 Gemini 4 Carbon 67.2% Oct 2026 Official leaderboard
9 Qwen4-Max 66.7% Nov 2026 Official leaderboard
10 Gemini 3.8 Flash 65.8% Sep 2026 Official leaderboard
11 Muse Wonder 65.5% Oct 2026 Official leaderboard
12 DeepSeek V4.1 Flash 65.4% Sep 2026 Official leaderboard
13 Claude Haiku 5.5 65.4% Oct 2026 Official leaderboard
14 Gemini 4 Hydrogen 65.1% Nov 2026 Official leaderboard
15 GLM-5.3 65% Aug 2026 Official leaderboard
16 Tencent Hy4 64.7% Nov 2026 Official leaderboard
17 MiMo-V2.6 Pro 64.3% Sep 2026 Official leaderboard
18 Muse Ember 63.8% Nov 2026 Official leaderboard
19 MiMo-V2.6 Flash 63.3% Sep 2026 Official leaderboard
20 DeepSeek V4.1 Flash 1017 62.9% Oct 2026 Official leaderboard
21 GPT-6.1 Sol 61.8% Sep 2026 Official leaderboard
22 Mistral Large 4 61.7% Oct 2026 Official leaderboard
23 Tencent Hy4 preview 61.6% Aug 2026 Official leaderboard
24 Kimi K3.1 Pro 61.5% Oct 2026 Official leaderboard
25 Mistral Large 4 Preview 61.4% Oct 2026 Official leaderboard
26 GPT-6.2 Astra 60.8% Nov 2026 Official leaderboard
27 Muse Spark Open 60.7% Oct 2026 Official leaderboard
28 GLM-5.3-Flash 60.6% Aug 2026 Official leaderboard
29 Grok 4.7 60.6% Sep 2026 Official leaderboard
30 GPT-6 Astra 60.4% Sep 2026 Official leaderboard
31 Kimi K3 60% Jul 2026 Official leaderboard
32 GPT-6.2 Sol 58.9% Nov 2026 Official leaderboard
33 Step 5 Preview 58.9% Sep 2026 Official leaderboard
34 Qwen3.8 Max (0902) 58.8% Sep 2026 Official leaderboard
35 MiniMax-M3.1-Flash 58.8% Oct 2026 Official leaderboard
36 DeepSeek V4 Pro 0813 57.1% Aug 2026 Official leaderboard
37 GPT-6 Luna 56.8% Sep 2026 Official leaderboard
38 Inkling 2 Mini 56.5% Nov 2026 Lab model card
39 GPT-6.1 Terra 56% Oct 2026 Official leaderboard
40 Mercury 3 54.7% Oct 2026 Official leaderboard
41 MiniMax-M3 54.3% Jun 2026 Official leaderboard
42 Beam 1.0 51.8% Oct 2026 Official leaderboard
43 Inkling 49.9% Jul 2026 Official leaderboard
44 Nemotron 3.5 Super VL 47.7% Oct 2026 Official leaderboard
45 Inkling-Small 44.8% Jul 2026 Official leaderboard

Labs report scores for 2 models in their cards. Switch the scores above to rank them together.

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% 50% 60% 70% 80% 101001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Claude Fable 5.2Muse EmberBeam 1.0Nemotron 3.5 Super VL 40% 50% 60% 70% 80% 101001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Claude Fable 5.2Muse EmberBeam 1.0Nemotron 3.5 Super VL
ModelSettingAccuracyLatency (mean per test)
Claude Fable 5.2 69.5% 207.62 s
Gemini 4 Neon 68.6% 1,232 s
Claude Opus 5.5 67.9% 1,980 s
Gemini 4 Argon 67.7% 793.03 s
Muse Wonder Full 67.6% 850.3 s
Inkling 2 Small 67.5% 296.33 s
Claude Sonnet 5.5 67.3% 2,066 s
Gemini 4 Carbon 67.2% 1,780 s
Qwen4-Max 66.7% 1,465 s
Gemini 3.8 Flash 65.8% 443.89 s
Muse Wonder 65.6% 310.13 s
DeepSeek V4.1 Flash 65.4% 960.25 s
Claude Haiku 5.5 65.4% 1,865 s
Gemini 4 Hydrogen 65.1% 1,078 s
GLM-5.3 65% 1,206 s
Tencent Hy4 64.7% 2,239 s
MiMo-V2.6 Pro 64.3% 682.01 s
Muse Ember 63.8% 89.35 s
MiMo-V2.6 Flash 63.3% 936.35 s
DeepSeek V4.1 Flash 1017 62.9% 909.21 s
GPT-6.1 Sol 61.8% 1,145 s
Mistral Large 4 61.7% 387.15 s
Tencent Hy4 preview 61.6% 1,503 s
Kimi K3.1 Pro 61.5% 2,099 s
Mistral Large 4 Preview 61.4% 986.07 s
Hy4 Flash 60.8% 344.74 s
GPT-6.2 Astra 60.8% 2,258 s
Muse Spark Open 60.7% 1,612 s
GLM-5.3-Flash 60.6% 296.1 s
Grok 4.7 60.6% 652.29 s
GPT-6 Astra 60.4% 1,229 s
Kimi K3 60% 875.04 s
GPT-6.2 Sol 58.9% 1,318 s
Step 5 Preview 58.9% 542.51 s
MiniMax-M3.1-Flash 58.8% 904.2 s
Qwen3.8 Max (0902) 58.8% 3,320 s
DeepSeek V4 Pro 0813 57.1% 476.99 s
GPT-6 Luna 56.8% 926.86 s
GPT-6.1 Terra 56% 1,787 s
Mercury 3 54.7% 232.76 s
MiniMax-M3 54.3% 189.32 s
Beam 1.0 51.8% 85.43 s
Inkling 49.9% 573.75 s
Nemotron 3.5 Super VL 47.7% 79.47 s
Inkling-Small 44.8% 314.22 s

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% 50% 60% 70% 80% 101001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Claude Fable 5.2Muse EmberBeam 1.0Nemotron 3.5 Super VL 40% 50% 60% 70% 80% 101001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Claude Fable 5.2Muse EmberBeam 1.0Nemotron 3.5 Super VL
ModelSettingAccuracyLatency (mean per test)
Claude Fable 5.2 69.5% 207.62 s
Gemini 4 Neon 68.6% 1,232 s
Claude Opus 5.5 67.9% 1,980 s
Gemini 4 Argon 67.7% 793.03 s
Muse Wonder Full 67.6% 850.3 s
Inkling 2 Small 67.5% 296.33 s
Claude Sonnet 5.5 67.3% 2,066 s
Gemini 4 Carbon 67.2% 1,780 s
Qwen4-Max 66.7% 1,465 s
Gemini 3.8 Flash 65.8% 443.89 s
Muse Wonder 65.6% 310.13 s
DeepSeek V4.1 Flash 65.4% 960.25 s
Claude Haiku 5.5 65.4% 1,865 s
Gemini 4 Hydrogen 65.1% 1,078 s
GLM-5.3 65% 1,206 s
Tencent Hy4 64.7% 2,239 s
MiMo-V2.6 Pro 64.3% 682.01 s
Muse Ember 63.8% 89.35 s
MiMo-V2.6 Flash 63.3% 936.35 s
DeepSeek V4.1 Flash 1017 62.9% 909.21 s
GPT-6.1 Sol 61.8% 1,145 s
Mistral Large 4 61.7% 387.15 s
Tencent Hy4 preview 61.6% 1,503 s
Kimi K3.1 Pro 61.5% 2,099 s
Mistral Large 4 Preview 61.4% 986.07 s
GPT-6.2 Astra 60.8% 2,258 s
Muse Spark Open 60.7% 1,612 s
GLM-5.3-Flash 60.6% 296.1 s
Grok 4.7 60.6% 652.29 s
GPT-6 Astra 60.4% 1,229 s
Kimi K3 60% 875.04 s
GPT-6.2 Sol 58.9% 1,318 s
Step 5 Preview 58.9% 542.51 s
MiniMax-M3.1-Flash 58.8% 904.2 s
Qwen3.8 Max (0902) 58.8% 3,320 s
DeepSeek V4 Pro 0813 57.1% 476.99 s
GPT-6 Luna 56.8% 926.86 s
GPT-6.1 Terra 56% 1,787 s
Mercury 3 54.7% 232.76 s
MiniMax-M3 54.3% 189.32 s
Beam 1.0 51.8% 85.43 s
Inkling 49.9% 573.75 s
Nemotron 3.5 Super VL 47.7% 79.47 s
Inkling-Small 44.8% 314.22 s

The best score, 69.5%, costs $4.83 a task

Runs on the cost frontier score higher than any run that's cheaper. Inkling 2 Small comes within 1.9 points of the best for $0.18 a task, 27 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind Meta Thinking Machines Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5.2 Inkling 2 Small Muse Ember 0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5.2 Inkling 2 Small

Source: Official Vals Finance Agent v2.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.2 69.5% $4.83 23,032
Gemini 4 Neon 68.6% $5.85 201,487
Claude Opus 5.5 67.8% $8.91 240,843
Gemini 4 Argon 67.7% $4.32 70,441
Muse Wonder Full 67.6% $2.68 40,520
Inkling 2 Small 67.5% $0.18 55,202
Claude Sonnet 5.5 67.3% $6.95 332,461
Gemini 4 Carbon 67.2% $7.29 171,664
Qwen4-Max 66.7% $1.74 118,951
Gemini 3.8 Flash 65.8% $1.99 58,676
Muse Wonder 65.5% $2.62 34,084
DeepSeek V4.1 Flash 65.4% $0.37 186,072
Claude Haiku 5.5 65.4% $1.64 459,542
Gemini 4 Hydrogen 65.1% $4.36 318,627
GLM-5.3 65% $1.03 82,474
Tencent Hy4 64.7% $1.28 246,830
MiMo-V2.6 Pro 64.3% $0.18 24,921
Muse Ember 63.8% $0.06 30,075
MiMo-V2.6 Flash 63.3% $0.07 32,508
DeepSeek V4.1 Flash 1017 62.9% $0.47 211,457
GPT-6.1 Sol 61.8% $1.63 48,985
Mistral Large 4 61.7% $0.35 38,442
Tencent Hy4 preview 61.6% $0.67 111,268
Kimi K3.1 Pro 61.5% $2.29 109,319
Mistral Large 4 Preview 61.4% $0.52 65,158
GPT-6.2 Astra 60.8% $7.98 66,546
Muse Spark Open 60.7% $2.76 31,778
GLM-5.3-Flash 60.6% $0.07 39,147
Grok 4.7 60.6% $2.24 50,375
GPT-6 Astra 60.4% $7.82 49,378
Kimi K3 60% $1.88 33,749
GPT-6.2 Sol 58.9% $2.06 60,089
Step 5 Preview 58.9% $0.58 45,002
Qwen3.8 Max (0902) 58.8% $2.99 161,507
MiniMax-M3.1-Flash 58.8% $0.16 121,024
DeepSeek V4 Pro 0813 57.1% $0.82 60,196
GPT-6 Luna 56.8% $0.12 102,259
GPT-6.1 Terra 56% $1.92 155,367
Mercury 3 54.7% $0.06 47,127
MiniMax-M3 54.3% $0.30 22,660
Beam 1.0 51.8% 33,555
Inkling 49.9% $0.89 59,596
Nemotron 3.5 Super VL 47.7% $0.11 30,828
Inkling-Small 44.8% $0.21 52,337

The best score, 69.5%, costs $4.83 a task

Runs on the cost frontier score higher than any run that's cheaper. Inkling 2 Small comes within 1.9 points of the best for $0.18 a task, 27 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind Meta Thinking Machines Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5.2 Inkling 2 Small Muse Ember 0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5.2 Inkling 2 Small

Source: Official Vals Finance Agent v2.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.2 69.5% $4.83 23,032
Gemini 4 Neon 68.6% $5.85 201,487
Claude Opus 5.5 67.8% $8.91 240,843
Gemini 4 Argon 67.7% $4.32 70,441
Muse Wonder Full 67.6% $2.68 40,520
Inkling 2 Small 67.5% $0.18 55,202
Claude Sonnet 5.5 67.3% $6.95 332,461
Gemini 4 Carbon 67.2% $7.29 171,664
Qwen4-Max 66.7% $1.74 118,951
Gemini 3.8 Flash 65.8% $1.99 58,676
Muse Wonder 65.5% $2.62 34,084
DeepSeek V4.1 Flash 65.4% $0.37 186,072
Claude Haiku 5.5 65.4% $1.64 459,542
Gemini 4 Hydrogen 65.1% $4.36 318,627
GLM-5.3 65% $1.03 82,474
Tencent Hy4 64.7% $1.28 246,830
MiMo-V2.6 Pro 64.3% $0.18 24,921
Muse Ember 63.8% $0.06 30,075
MiMo-V2.6 Flash 63.3% $0.07 32,508
DeepSeek V4.1 Flash 1017 62.9% $0.47 211,457
GPT-6.1 Sol 61.8% $1.63 48,985
Mistral Large 4 61.7% $0.35 38,442
Tencent Hy4 preview 61.6% $0.67 111,268
Kimi K3.1 Pro 61.5% $2.29 109,319
Mistral Large 4 Preview 61.4% $0.52 65,158
GPT-6.2 Astra 60.8% $7.98 66,546
Muse Spark Open 60.7% $2.76 31,778
GLM-5.3-Flash 60.6% $0.07 39,147
Grok 4.7 60.6% $2.24 50,375
GPT-6 Astra 60.4% $7.82 49,378
Kimi K3 60% $1.88 33,749
GPT-6.2 Sol 58.9% $2.06 60,089
Step 5 Preview 58.9% $0.58 45,002
Qwen3.8 Max (0902) 58.8% $2.99 161,507
MiniMax-M3.1-Flash 58.8% $0.16 121,024
DeepSeek V4 Pro 0813 57.1% $0.82 60,196
GPT-6 Luna 56.8% $0.12 102,259
GPT-6.1 Terra 56% $1.92 155,367
Mercury 3 54.7% $0.06 47,127
MiniMax-M3 54.3% $0.30 22,660
Beam 1.0 51.8% 33,555
Inkling 49.9% $0.89 59,596
Nemotron 3.5 Super VL 47.7% $0.11 30,828
Inkling-Small 44.8% $0.21 52,337

2 models have held number one on the official leaderboard since its release in October 2026 2 models have held number one with claims included since its release in October 2026

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

Anthropic
Own leaderboard 2 leaders

Source: Official Vals Finance Agent v2.1 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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Fable 5.2 Gemini 3.8 Flash Kimi K3 MiniMax-M3 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Fable 5.2 Kimi K3

Sources: Official Vals Finance Agent v2.1 leaderboard and labs’ model cards

One lab put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Thinking Machines
DateLaunch postScoreAs the post writes it
Nov 5, 2026 Inkling 2 Small 67.5 Finance Agent v2.1, 67.5%

One system card reports it

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Nov 20, 2026 Inkling 2 Mini model card 56.5 Finance Agent v2.1 Capabilities

One lab claim next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 1 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Inkling 2 Small 67.5% Finance Agent v2.1, 67.5% 67.5% Leaderboard Same

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.

Anthropic OpenAI Google DeepMind Meta
Vals Finance Agent v2.1 18 leaders
Vals Finance Agent v1 Replaced by Agent, Feb 2026
Vals Finance Agent Replaced by v2, May 2026

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 12 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (45 entries), launch posts (1 score) and system cards (1 score).
  • Pass Models scored 45 models we track, plus 1 other entry on its leaderboard.
  • Pass Newest score The newest is from November 2026, for Inkling 2 Mini.
  • Pass Frontier models 9 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 69.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on vals.ai: 44 rows for models we track and 1 other entry at the last read.
  • Pass Last read November 19th, 2026.
  • Pass Read worked The last read worked.
Consistency 2 pass
  • Pass Lab claims The one pair agrees within 10 points.
  • Pass Known issues No published audit found problems.

How we check