Skip to content

Benchmarks · Vals AI

TaxEval v2

Measures answering tax questions (questions and answers created by Vals). Vals marks the leaderboard "Archived Benchmark": "Since performance on this benchmark has saturated, we no longer run this benchmark on new model releases" (last update 1 September 2026).

Models measured
108
Top
80.38 %Muse Spark 1.2within the standard error of ±0.76
Version
2
Measured
1 September 2026
Go to the sourceThis benchmark does not count towards the PrudentScore.

01

Ranking

At the top: Top 15

Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.

  1. 1Muse Spark 1.280.38%
  2. 2Muse Spark 1.179.72%
  3. 3Muse Spark77.68%
  4. 4Claude Sonnet 4.677.11%
  5. 5Claude Fable 576.94%
  6. 6=GPT-5.6 Luna76.17%
  7. 6=GPT-5.6 Terra76.17%
  8. 8=Claude Fable 5.175.96%
  9. 8=Claude Opus 4.6 (Thinking)75.96%
  10. 10GPT 5.275.76%
  11. 11Kimi K375.72%
  12. 12Grok 4 Fast (Reasoning)75.70%
  13. 13=Claude Opus 4.875.63%
  14. 13=Claude Sonnet 575.63%
  15. 15GLM 5.3 Flash75.59%

Source: Vals AI (TaxEval v2) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.

Passes the compliance gateRejected by the compliance gate
TaxEval v2 2
# ModelScoreCost/taskGatePrudentScore
1Muse Spark 1.2Meta80.38%± 0.76-Rejectedno score
2Muse Spark 1.1Meta79.72%± 0.77-Rejectedno score
3Muse SparkMeta77.68%± 0.81-Rejectedno score
4Claude Sonnet 4.6Anthropic77.11%± 0.82-Passes55
5Claude Fable 5Anthropic76.94%± 0.82-Rejectedno score
6=GPT-5.6 LunaOpenAI76.17%± 0.84-Passes56
6=GPT-5.6 TerraOpenAI76.17%± 0.84-Passes53
8=Claude Fable 5.1Anthropic75.96%± 0.83-Rejectedno score
8=Claude Opus 4.6 (Thinking)Anthropic75.96%± 0.83-Passesno score
10GPT 5.2OpenAI75.76%± 0.84-Passesno score
11Kimi K3Moonshot AI75.72%± 0.84-Rejectedno score
12Grok 4 Fast (Reasoning)xAI (SpaceXAI)75.70%± 0.76-Rejectedno score
13=Claude Opus 4.8Anthropic75.63%± 0.84-Passes62
13=Claude Sonnet 5Anthropic75.63%± 0.84-Passes59
15GLM 5.3 FlashZ.ai75.59%± 0.84-Rejectedno score
16Qwen 3.8 MaxAlibaba75.55%± 0.84-Not yet assessedno score
17Inkling SmallThinking Machines75.51%± 0.84-Rejectedno score
18=InklingThinking Machines75.31%± 0.84-Rejectedno score
18=Qwen 3.7 MaxAlibaba75.31%± 0.84-Not yet assessedno score
20Claude Opus 4.7Anthropic75.27%± 0.84-Passes54
21GPT 5 MiniOpenAI75.22%± 0.85-Passesno score
22Claude Opus 5Anthropic75.14%± 0.83-Passes72
23GPT 4.1OpenAI75.06%± 0.84-Passesno score
24GPT 5.5OpenAI74.98%± 0.85-Passes51
25=Gemini 3.6 FlashGoogle74.86%± 0.85-Passes46
25=GPT 5.1OpenAI74.86%± 0.85-Passesno score
27Claude Opus 4.5 (Thinking)Anthropic74.86%± 0.85-Passesno score
28o4 MiniOpenAI74.78%± 0.85-Passesno score
29GPT-5.6 SolOpenAI74.78%± 0.86-Passes57
30=Gemini 3.7 FlashGoogle74.73%± 0.85-Passes58
30=Qwen 3.6 PlusAlibaba74.73%± 0.85-Not yet assessedno score
32Kimi K2.6Moonshot AI74.65%± 0.85-Rejectedno score
33o3OpenAI74.57%± 0.85-Passesno score
34GPT 4o (2024-11-20)OpenAI74.53%± 0.86-Passesno score
35Gemini 3.8 FlashGoogle74.45%± 0.85-Passes62
36Gemini 3.5 FlashGoogle74.37%± 0.85-Passes47
37Claude Opus 4.5 (Nonthinking)Anthropic74.33%± 0.85-Passesno score
38Kimi K2.5Moonshot AI74.20%± 0.85-Passes37
39Grok 4.20 (Reasoning)xAI (SpaceXAI)74.12%± 0.86-Rejectedno score
40=GPT 5.4 (xhigh)OpenAI73.96%± 0.87-Passes0
40=Qwen 3 Max PreviewAlibaba73.96%± 0.86-Not yet assessedno score
42Gemini 3 Flash (12/25)Google73.88%± 0.86-Rejectedno score
43MiMo V2.5 ProXiaomi73.79%± 0.86-Rejectedno score
44Claude Opus 4.1 (Thinking)Anthropic73.67%± 0.86-Rejectedno score
45Qwen 3 MaxAlibaba73.51%± 0.86-Not yet assessedno score
46GPT 5OpenAI73.39%± 0.87-Passesno score
47GLM 5.2Z.ai73.34%± 0.86-Rejectedno score
48Claude Sonnet 4.5 (Thinking)Anthropic73.30%± 0.86-Passesno score
49Grok 4.1 Fast (Reasoning)xAI (SpaceXAI)73.14%± 0.86-Rejectedno score
50Nemotron 3 UltraNVIDIA73.10%± 0.87-Rejectedno score
51=DeepSeek V4 Pro (0813)DeepSeek73.06%± 0.87-Rejectedno score
51=Mistral Large 3Mistral AI73.06%± 0.87-Passesno score
53Gemini 3.1 Pro Preview (02/26)Google72.88%± 0.86-Rejectedno score
54=Gemini 2.5 Flash Preview (9/25) (Nonthinking)Google72.73%± 0.87-Rejectedno score
54=MiniMax-M3MiniMax72.73%± 0.86-Rejectedno score
56Gemini 3.5 Flash LiteGoogle72.61%± 0.88-Passes38
57Gemini 3 Pro (11/25)Google72.57%± 0.87-Rejectedno score
58=Gemini 2.5 Flash Preview (9/25) (Thinking)Google72.40%± 0.87-Rejectedno score
58=GLM 4.5Z.ai72.40%± 0.87-Rejectedno score
60GLM 5.3Z.ai72.36%± 0.88-Rejectedno score
61Qwen 3.5 FlashAlibaba72.16%± 0.88-Not yet assessedno score
62DeepSeek V4 Pro (2026-04-23)DeepSeek72.08%± 0.88-Rejectedno score
63Claude Sonnet 4 (Thinking)Anthropic72.00%± 0.88-Rejectedno score
64=Claude Opus 4 (Nonthinking)Anthropic71.91%± 0.88-Rejectedno score
64=GPT 4.1 MiniOpenAI71.91%± 0.88-Passesno score
66MiMo V2.5Xiaomi71.83%± 0.88-Rejectedno score
67Gemini 3.1 Flash Lite PreviewGoogle71.79%± 0.88-Not yet assessedno score
68Kimi K2 ThinkingMoonshot AI71.71%± 0.88-Rejectedno score
69Grok 4.5xAI (SpaceXAI)71.67%± 0.89-Rejectedno score
70GPT OSS 120BOpenAI71.59%± 0.88-Passesno score
71Grok 4 Fast (Non-Reasoning)xAI (SpaceXAI)71.58%± 0.78-Rejectedno score
72Claude Opus 4.1 (Nonthinking)Anthropic71.46%± 0.88-Rejectedno score
73Qwen 3.6 27BAlibaba71.26%± 0.88-Rejectedno score
74GPT 5.4 MiniOpenAI71.22%± 0.90-Passes25
75GLM 5.1Z.ai71.19%± 0.90-Rejectedno score
76Grok 4.6xAI (SpaceXAI)71.10%± 0.90-Rejectedno score
77Qwen 3.8 27BAlibaba70.85%± 0.89-Rejectedno score
78Grok 4.3xAI (SpaceXAI)70.81%± 0.90-Rejectedno score
79DeepSeek V4 Flash 0731DeepSeek70.69%± 0.90-Rejectedno score
80Ling 3.0 FlashAnt Group70.65%± 0.90-Rejectedno score
81Mistral Medium 3.1 (05/2025)Mistral AI70.32%± 0.89-Passesno score
82Kimi K2 InstructMoonshot AI70.20%± 0.90-Rejectedno score
83GLM 5Z.ai70.03%± 0.90-Passesno score
84Claude Sonnet 4 (Nonthinking)Anthropic69.62%± 0.90-Rejectedno score
85GLM 4.7Z.ai68.77%± 0.92-Passesno score
86=DeepSeek V3.2 (Thinking)DeepSeek68.15%± 0.92-Passesno score
86=MiniMax-M2.5MiniMax68.15%± 0.91-Passesno score
88Mistral Medium 3.5Mistral AI67.99%± 0.92-Passes23
89Claude Haiku 4.5 (Thinking)Anthropic67.54%± 0.91-Passes32
90GPT 5.4 NanoOpenAI67.42%± 0.93-Passes29
91GPT 5 NanoOpenAI67.38%± 0.91-Passesno score
92=Llama 4 MaverickMeta66.56%± 0.92-Rejectedno score
92=MiniMax-M2.7MiniMax66.56%± 0.92-Rejectedno score
94MiniMax-M2.1MiniMax66.35%± 0.93-Passesno score
95GLM 4.6Z.ai66.23%± 0.81-Rejectedno score
96Gemini 2.5 Flash Lite (9/25) (Nonthinking)Google66.23%± 0.92-Rejectedno score
97Grok 4xAI (SpaceXAI)65.09%± 0.96-Rejectedno score
98Gemini 2.5 Flash Lite (9/25) (Thinking)Google64.72%± 0.93-Rejectedno score
99GPT OSS 20BOpenAI63.70%± 0.94-Passesno score
100Magistral Medium 1.2 (09/2025)Mistral AI61.94%± 0.96-Not yet assessedno score
101Grok 4.1 Fast Non-ReasoningxAI (SpaceXAI)61.86%± 0.91-Rejectedno score
102Command ACohere61.37%± 0.95-Rejectedno score
103GPT 4.1 NanoOpenAI60.75%± 0.96-Passesno score
104Magistral Small 1.2 (09/2025)Mistral AI60.30%± 0.97-Passesno score
105Laguna XS.2Poolside58.95%± 0.96-Rejectedno score
106Mistral Small 3.1 (03/2025)Mistral AI58.30%± 0.97-Passesno score
107Llama 4 ScoutMeta55.19%± 0.97-Rejectedno score
108Laguna M.1Poolside1.64%± 0.26-Rejectedno score

Source: Vals AI (TaxEval v2) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.

Vals AI · Version 2 · Measured 1 September 2026 · Go to the source. The PrudentScore is a composition by Prudai B.V. based on publicly published figures and does not represent the view or endorsement of Vals AI, Harvey or Artificial Analysis. Figures from Artificial Analysis appear in running text only, with attribution and a link, as that party's terms require [9].

Other benchmarks

Claude and Anthropic, GPT and OpenAI, Gemini and Google, Grok and xAI, Kimi and Moonshot AI, Muse and Meta, Mistral, DeepSeek, GLM and Z.ai, Harvey, Vals AI and Artificial Analysis are trademarks of their respective owners. They are named for identification only and this implies no partnership, sponsorship or endorsement.