Share promptsSign inCreate team library
← All benchmarks

Finance AI benchmarks

Analyst research on SEC filings and long credit agreements. 2 public leaderboards below.

Finance Agent

Agents do entry-level financial analyst research on SEC filings (retrieval, calculations, beat/miss, modelling).

Run by Vals AI. Leaderboard as of 19 Oct 2026 · retrieved 21 Oct 2026. Showing top 25 of 28 models listed.

  • Anthropic
  • OpenAI
  • Google
  • xAI
  • Meta
  • DeepSeek
  • Z.ai
  • Moonshot
  • MiniMax
  • Mistral
  • Xiaomi
  • Other
Finance Agent: Accuracy vs Cost per test (USD)Scatter plot of 25 models. Leader on Accuracy: Claude Opus 5.5 at 67.85%. Full values in the table below.$.01$.1$1$1050%55%60%65%70%Cost per test (USD) (log scale) →↑ AccuracyMiniMax-M3 (MiniMax): Cost per test (USD) $0.30, Accuracy 54.34%MMGPT-6 Luna (OpenAI): Cost per test (USD) $0.12, Accuracy 56.84%ODeepSeek V4 Pro 0813 (DeepSeek): Cost per test (USD) $0.82, Accuracy 57.12%DMiniMax-M3.1-Flash (MiniMax): Cost per test (USD) $0.16, Accuracy 58.83%MMStep 5 Preview (Other): Cost per test (USD) $0.58, Accuracy 58.90%·Kimi K3 (Moonshot): Cost per test (USD) $1.88, Accuracy 60.01%KGPT-6 Astra (OpenAI): Cost per test (USD) $7.82, Accuracy 60.43%OGrok 4.7 (xAI): Cost per test (USD) $2.24, Accuracy 60.59%XGLM 5.3 Flash (Z.ai): Cost per test (USD) $0.070, Accuracy 60.59%ZMuse Spark Open (Meta): Cost per test (USD) $2.76, Accuracy 60.66%MMistral Large 4 (Mistral): Cost per test (USD) $1.02, Accuracy 61.38%MiHy4 Preview (Other): Cost per test (USD) $0.67, Accuracy 61.62%·GPT-6.1 Sol (OpenAI): Cost per test (USD) $1.63, Accuracy 61.82%ODeepSeek V4.1 Flash 1017 (DeepSeek): Cost per test (USD) $0.47, Accuracy 62.93%DMiMo V2.6 Flash (Xiaomi): Cost per test (USD) $0.070, Accuracy 63.32%XiMiMo V2.6 Pro (Xiaomi): Cost per test (USD) $0.18, Accuracy 64.26%XiGLM 5.3 (Z.ai): Cost per test (USD) $1.03, Accuracy 65.00%ZClaude Haiku 5.5 (Anthropic): Cost per test (USD) $1.64, Accuracy 65.37%ADeepSeek V4.1 Flash (DeepSeek): Cost per test (USD) $0.37, Accuracy 65.43%DWonder (Meta): Cost per test (USD) $2.62, Accuracy 65.55%MGemini 3.8 Flash (Google): Cost per test (USD) $1.99, Accuracy 65.76%GClaude Sonnet 5.5 (Anthropic): Cost per test (USD) $6.95, Accuracy 67.28%AWonder Full (Meta): Cost per test (USD) $2.68, Accuracy 67.58%MGemini 4 Argon (Google): Cost per test (USD) $4.32, Accuracy 67.66%GClaude Opus 5.5 (Anthropic): Cost per test (USD) $8.91, Accuracy 67.85%AClaude Opus 5.5Gemini 4 ArgonWonder Full
Finance Agent: top 15 by AccuracyRanked bars. Leader: Claude Opus 5.5 at 67.85%. Full values in the table below.0%20%40%60%80%100%Claude Opus 5.5 (Anthropic): 67.85%AClaude Opus 5.567.85%Gemini 4 Argon (Google): 67.66%GGemini 4 Argon67.66%Wonder Full (Meta): 67.58%MWonder Full67.58%Claude Sonnet 5.5 (Anthropic): 67.28%AClaude Sonnet 5.567.28%Gemini 3.8 Flash (Google): 65.76%GGemini 3.8 Flash65.76%Wonder (Meta): 65.55%MWonder65.55%DeepSeek V4.1 Flash (DeepSeek): 65.43%DDeepSeek V4.1 Flash65.43%Claude Haiku 5.5 (Anthropic): 65.37%AClaude Haiku 5.565.37%GLM 5.3 (Z.ai): 65.00%ZGLM 5.365.00%MiMo V2.6 Pro (Xiaomi): 64.26%XiMiMo V2.6 Pro64.26%MiMo V2.6 Flash (Xiaomi): 63.32%XiMiMo V2.6 Flash63.32%DeepSeek V4.1 Flash 1017 (DeepSeek): 62.93%DDeepSeek V4.1 Flash 101762.93%GPT-6.1 Sol (OpenAI): 61.82%OGPT-6.1 Sol61.82%Hy4 Preview (Other): 61.62%·Hy4 Preview61.62%Mistral Large 4 (Mistral): 61.38%MiMistral Large 461.38%
Accuracy
Share of analyst questions with a correct final answer.
Cost per test (USD)
Average API cost to run one task, as published.
Finance Agent scores
RankModelProviderAccuracyCost per test (USD)Price in / out ($/1M tokens)
1 Claude Opus 5.5Anthropic67.85%$8.91$4 / $20
2 Gemini 4 ArgonGoogle67.66%$4.32$4 / $20
3 Wonder FullMeta67.58%$2.68$4 / $20
4 Claude Sonnet 5.5Anthropic67.28%$6.95$2 / $10
5 Gemini 3.8 FlashGoogle65.76%$1.99$1.5 / $7.5
6 WonderMeta65.55%$2.62$4 / $20
7 DeepSeek V4.1 FlashDeepSeek65.43%$0.37$0.3 / $1.2
8 Claude Haiku 5.5Anthropic65.37%$1.64$0.1 / $0.5
9 GLM 5.3Z.ai65.00%$1.03$1.4 / $4.4
10 MiMo V2.6 ProXiaomi64.26%$0.18$0.435 / $0.87
Show 15 more rows
RankModelProviderAccuracyCost per test (USD)Price in / out ($/1M tokens)
11 MiMo V2.6 FlashXiaomi63.32%$0.070$0.14 / $0.28
12 DeepSeek V4.1 Flash 1017DeepSeek62.93%$0.47$0.3 / $1.2
13 GPT-6.1 SolOpenAI61.82%$1.63$2 / $10
14 Hy4 PreviewOther61.62%$0.67$0.834 / $2.501
15 Mistral Large 4Mistral61.38%$1.02$1.36 / $4.18
16 Muse Spark OpenMeta60.66%$2.76$1.25 / $4.25
17 GLM 5.3 FlashZ.ai60.59%$0.070$0.075 / $0.25
18 Grok 4.7xAI60.59%$2.24$2 / $6
19 GPT-6 AstraOpenAI60.43%$7.82$10 / $50
20 Kimi K3Moonshot60.01%$1.88$3 / $15
21 Step 5 PreviewOther58.90%$0.58$1 / $2.7
22 MiniMax-M3.1-FlashMiniMax58.83%$0.16$0.15 / $0.6
23 DeepSeek V4 Pro 0813DeepSeek57.12%$0.82$1.32 / $3.96
24 GPT-6 LunaOpenAI56.84%$0.12$0.1 / $0.5
25 MiniMax-M3MiniMax54.34%$0.30$0.6 / $2.4

Source: Vals AI leaderboard ↗

Top model here: Claude Opus 5.5 (Anthropic). Try prompts for Claude →

CorpFin v2

Questions about long credit agreements, testing understanding of long financial documents.

Run by Vals AI. Leaderboard as of 12 Aug 2026 · retrieved 21 Oct 2026. Showing top 25 of 134 models listed.

  • Anthropic
  • OpenAI
  • Google
  • xAI
  • Meta
  • Alibaba
  • Z.ai
  • Moonshot
  • MiniMax
  • Other
CorpFin v2: top 15 by AccuracyRanked bars. Leader: Claude Opus 5 at 73.19%. Full values in the table below.0%20%40%60%80%100%Claude Opus 5 (Anthropic): 73.19%AClaude Opus 573.19%Claude Fable 5 (Anthropic): 71.83%AClaude Fable 571.83%Kimi K3 (Moonshot): 71.56%KKimi K371.56%Muse Spark 1.1 (Meta): 71.29%MMuse Spark 1.171.29%Muse Spark 1.2 (Meta): 70.94%MMuse Spark 1.270.94%Inkling Small (Other): 69.62%·Inkling Small69.62%Inkling (Other): 68.57%·Inkling68.57%Grok 4.3 (xAI): 68.53%XGrok 4.368.53%GPT 5.5 (OpenAI): 68.42%OGPT 5.568.42%Kimi K2.5 (Moonshot): 68.26%KKimi K2.568.26%MiniMax-M3 (MiniMax): 68.10%MMMiniMax-M368.10%Qwen 3 Max Thinking (Alibaba): 68.03%QQwen 3 Max Thinking68.03%Grok 4.5 (xAI): 67.41%XGrok 4.567.41%Claude Opus 4.6 (Thinking) (Anthropic): 67.02%AClaude Opus 4.6 (Thinking)67.02%Claude Sonnet 5 (Anthropic): 66.98%AClaude Sonnet 566.98%
Accuracy
Share of questions answered correctly.
CorpFin v2 scores
RankModelProviderAccuracyPrice in / out ($/1M tokens)
1 Claude Opus 5Anthropic73.19%$5 / $25
2 Claude Fable 5Anthropic71.83%$10 / $50
3 Kimi K3Moonshot71.56%$3 / $15
4 Muse Spark 1.1Meta71.29%$1.25 / $4.25
5 Muse Spark 1.2Meta70.94%$1.25 / $4.25
6 Inkling SmallOther69.62%$0.3 / $1.2
7 InklingOther68.57%$1 / $4.05
8 Grok 4.3xAI68.53%$1.25 / $2.5
9 GPT 5.5OpenAI68.42%$5 / $30
10 Kimi K2.5Moonshot68.26%$0.6 / $3
Show 15 more rows
RankModelProviderAccuracyPrice in / out ($/1M tokens)
11 MiniMax-M3MiniMax68.10%$0.6 / $2.4
12 Qwen 3 Max ThinkingAlibaba68.03%$1.2 / $6
13 Grok 4.5xAI67.41%$2 / $6
14 Claude Opus 4.6 (Thinking)Anthropic67.02%$5 / $25
15 Claude Sonnet 5Anthropic66.98%$2 / $10
16 Grok 4 Fast (Reasoning)xAI66.90%$0.2 / $0.5
17 Kimi K2.6Moonshot66.74%$0.95 / $4
18 Claude Opus 4.8Anthropic66.71%$5 / $25
19 Qwen 3.6 Max PreviewAlibaba66.47%$1.3 / $7.8
20 Gemini 3 Flash (12/25)Google66.43%$0.5 / $3
21 Grok 4.6xAI66.16%$2 / $6
22 GLM 5.2Z.ai66.12%$1.4 / $4.4
23 Claude Opus 4.7Anthropic66.08%$5 / $25
24 Grok 4xAI66.05%$3 / $15
25 Grok 4.1 Fast (Reasoning)xAI65.97%$0.2 / $0.5

Source: Vals AI leaderboard ↗

Top model here: Claude Opus 5 (Anthropic). Try prompts for Claude →

ChatJudge does not run these tests. Scores are copied as published by each source, with a link and the date we read them, and only change when we refresh from the source. Different leaderboards use different settings, so compare models within one chart, not across charts. Spotted a mistake? Tell us.