Finance Agent v2
We show this table for reference; we do not rank on it.
Gemini 4 Argon leads Vals' October 20, 2026 Finance Agent v2 table at 65.4% partial credit. Its all-pass score is 55.8%. The test checks financial research with tools; partial credit rewards weighted checks after required dealbreakers pass, while all-pass requires every check.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
View the latest Finance Agent v2.1 rankingsFinance Agent v2 score on Finance Agent v2 — October 20, 2026
We mirror the published finance agent v2 score view for Finance Agent v2. Gemini 4 Argon leads the public snapshot at 65.4%, followed by Gemini 3.8 Flash (61.4%) and Muse Spark Open (61.3%). We do not use these results to rank models overall.
Gemini 4 Argon
high reasoning
Gemini 3.8 Flash
high reasoning
Muse Spark Open
Meta
xhigh reasoning
81 modelsAgenticCurrentDisplay onlyUpdated October 20, 2026
Finance Agent v2 score table (81 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Gemini 4 ArgonGooglehigh reasoning | 65.4% | Not reported | Closed |
| 2 | Gemini 3.8 FlashGooglehigh reasoning | 61.4% | Not reported | Closed |
| 3 | Muse Spark OpenMetaxhigh reasoning | 61.3% | Not reported | Open |
| 4 | Muse Spark 1.2Metaxhigh reasoning | 60.6% | Not reported | Closed |
| 5 | Muse Spark 1.3 MaxMetamax reasoning | 60.0% | Not reported | Unknown |
| 6 | Muse Wonder FullMetamax reasoning | 59.1% | Not reported | Closed |
| 7 | Gemini 3.7 FlashGooglehigh reasoning | 59.0% | Not reported | Closed |
| 8 | Muse Spark 1.3Metaxhigh reasoning | 58.9% | Not reported | Closed |
| 9 | Claude Fable 5.1Anthropic | 58.9% | Not reported | Closed |
| 10 | Claude Opus 5Anthropic | 58.6% | Not reported | Closed |
| 11 | Claude Opus 5.5Anthropic | 58.6% | Not reported | Closed |
| 12 | Claude Sonnet 5.5Anthropic | 58.1% | Not reported | Closed |
| 13 | Qwen3.8 Max InternalAlibaba | 57.9% | Not reported | Unknown |
| 14 | Gemini 3.5 FlashGooglehigh reasoning | 57.9% | Not reported | Closed |
| 15 | GLM-5.3-FlashZ.AImax reasoning | 57.9% | Not reported | Open |
| 16 | MiMo-V2.6-ProXiaomi | 57.3% | Not reported | Open |
| 17 | Muse Spark 1.1Metaxhigh reasoning | 57.2% | Not reported | Closed |
| 18 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 57.0% | Not reported | Open |
| 19 | Muse WonderMetamax reasoning | 56.6% | Not reported | Closed |
| 20 | Claude Fable 5Anthropic | 56.3% | Not reported | Closed |
| 21 | Gemini 3.6 FlashGooglehigh reasoning | 56.3% | Not reported | Closed |
| 22 | MiMo-V2.6-FlashXiaomi | 56.3% | Not reported | Open |
| 23 | GLM-5.3Z.AImax reasoning | 55.8% | Not reported | Open |
| 24 | Hy4 previewTencent | 55.1% | Not reported | Open |
| 25 | GPT-5.6 LunaOpenAImax reasoning | 55.0% | Not reported | Closed |
| 26 | Ling 3.0 Flash Af Rc3Ant | 54.9% | Not reported | Unknown |
| 27 | Mistral Large 4Mistralhigh reasoning | 54.7% | Not reported | Open |
| 28 | GPT-5.6 TerraOpenAImax reasoning | 54.4% | Not reported | Closed |
| 29 | Claude Haiku 5.5Anthropic | 54.1% | Not reported | Closed |
| 30 | Claude Opus 4.8Anthropic | 53.9% | Not reported | Closed |
| 31 | Claude Sonnet 5Anthropic | 53.9% | Not reported | Closed |
| 32 | GPT-5.6 SolOpenAImax reasoning | 53.8% | Not reported | Closed |
| 33 | Grok 4.6xAIhigh reasoning | 53.7% | Not reported | Closed |
| 34 | GPT-6 AstraOpenAImax reasoning | 53.5% | Not reported | Closed |
| 35 | DeepSeek V4.1 Flash 1017DeepSeekhigh reasoning | 53.5% | Not reported | Open |
| 36 | Kimi K3Moonshot AImax reasoning | 53.1% | Not reported | Open |
| 37 | Grok 4.7xAIxhigh reasoning | 52.3% | Not reported | Closed |
| 38 | GPT-6.1 SolOpenAImax reasoning | 52.0% | Not reported | Closed |
| 39 | Ember-1Fireworks | 51.9% | Not reported | Closed |
| 40 | GPT-5.5OpenAIxhigh reasoning | 51.8% | Not reported | Closed |
| 41 | Claude Opus 4.7Anthropic | 51.5% | Not reported | Closed |
| 42 | Claude Sonnet 4.6Anthropic | 51.0% | Not reported | Closed |
| 43 | Step 5 PreviewStepFun | 50.7% | Not reported | Open |
| 44 | Qwen3.8 MaxAlibaba | 50.6% | Not reported | Open |
| 45 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 50.4% | Not reported | Open |
| 46 | GPT-6 LunaOpenAImax reasoning | 49.9% | Not reported | Closed |
| 47 | GLM-5.2Z.AI | 49.7% | Not reported | Open |
| 48 | DeepSeek V4 Flash 0731DeepSeekhigh reasoning | 49.5% | Not reported | Open |
| 49 | GPT-6 SolOpenAImax reasoning | 49.0% | Not reported | Closed |
| 50 | Qwen3.8-27BAlibabaxhigh reasoning | 48.6% | Not reported | Open |
| 51 | Grok 4.5xAIhigh reasoning | 48.3% | Not reported | Closed |
| 52 | MiniMax M3MiniMax | 48.3% | Not reported | Open |
| 53 | Qwen3.7 MaxAlibaba | 47.8% | Not reported | Closed |
| 54 | Gemini 3.5 Flash-LiteGooglehigh reasoning | 47.4% | Not reported | Closed |
| 55 | InklingThinking Machines Lab0.99 reasoning | 46.6% | Not reported | Open |
| 56 | GPT-5.4 miniOpenAIxhigh reasoning | 45.4% | Not reported | Closed |
| 57 | Kimi K2.6Moonshot AI | 44.9% | Not reported | Open |
| 58 | GLM-5.1Z.AI | 44.8% | Not reported | Open |
| 59 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 44.1% | Not reported | Open |
| 60 | Gemini 3.1 Pro PreviewGooglehigh reasoning | 43.0% | Not reported | Unknown |
| 61 | Gemini 3 Flash PreviewGooglehigh reasoning | 42.6% | Not reported | Unknown |
| 62 | MiMo-V2.5-ProXiaomi | 41.5% | Not reported | Closed |
| 63 | Inkling-SmallThinking Machines Lab0.99 reasoning | 41.3% | Not reported | Open |
| 64 | Qwen3.6 PlusAlibaba | 40.8% | Not reported | Closed |
| 65 | Qwen3.7 PlusAlibaba | 38.2% | Not reported | Closed |
| 66 | GPT-5.4 nanoOpenAIhigh reasoning | 38.2% | Not reported | Closed |
| 67 | Grok 4.3xAIhigh reasoning | 37.7% | Not reported | Closed |
| 68 | Nemotron 3 Ultra 550b A55bNvidia | 37.7% | Not reported | Unknown |
| 69 | MiMo-V2.5Xiaomi | 36.7% | Not reported | Closed |
| 70 | Kimi K2.5 ThinkingMoonshot AI | 35.8% | Not reported | Unknown |
| 71 | Mistral Medium 3.5Mistral AIhigh reasoning | 32.1% | Not reported | Unknown |
| 72 | Claude Haiku 4.5 ThinkingAnthropic | 31.0% | Not reported | Closed |
| 73 | Ling 3.0 Flash 2607Ant | 30.3% | Not reported | Unknown |
| 74 | Gemini 3.1 Flash Lite PreviewGooglehigh reasoning | 30.0% | Not reported | Unknown |
| 75 | Grok 4.20 0309 ReasoningxAI | 28.5% | Not reported | Unknown |
| 76 | MiniMax M2.7MiniMax | 27.9% | Not reported | Open |
| 77 | Laguna M.1Poolside | 25.0% | Not reported | Closed |
| 78 | Mercury 2.5Inceptionhigh reasoning | 18.9% | Not reported | Closed |
| 79 | Nemotron Lightning 3p5 30b A3bFireworks AI | 18.5% | Not reported | Unknown |
| 80 | Laguna XS.2Poolside | 15.6% | Not reported | Open |
| 81 | Command A Plus 05 2026Cohere | 9.0% | Not reported | Unknown |
How Finance Agent v2 is shown here
BenchLM mirrors the public Vals AI Finance Agent v2 leaderboard captured from https://www.vals.ai/benchmarks/fabv2 and updated by Vals on October 20, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.
Finance Agent v2 is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.
Snapshot
The published Finance Agent v2 snapshot places Gemini 4 Argon first at 65.4%. The third row is 4.1 points behind. The broader top-10 range is 6.8 points, so many of the published results sit in a relatively narrow band.
81 models have been evaluated on Finance Agent v2. The benchmark falls in the Agentic category. Finance Agent v2 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Finance Agent v2
Year
2026
Tasks
Financial analyst task categories
Format
Mean score across repeated runs
Difficulty
Professional expert-task agent workflow
Archived Finance Agent v2 reports dealbreaker-gated partial credit, with all-pass as a separate metric. Finance Agent v2.1 changes the questions, rubrics, jury, and handling of hedged answers. Scores are not comparable between editions. This table stays display only.
Freshness and provenance
Version
Finance Agent v2 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Finance Agent v2 measure?
Vals AI benchmark for realistic financial analyst agent tasks across qualitative analysis, quantitative analysis, market work, comparables, precedents, earnings, disclosure, and modeling.
Which model leads the published Finance Agent v2 snapshot?
Gemini 4 Argon currently leads the published Finance Agent v2 snapshot with 65.4% finance agent v2 score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Finance Agent v2?
The October 20, 2026 snapshot contains 81 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.