Benchmarks / Agentic

Tax Agent Bench

Evaluating agents on research-grade US tax questions using the Tax Agent Bench harness

68models
–Open Index weight
2026-10-12last updated
Vals AIpublisher

Tax Agent Bench

Higher is better · best effort per model

Chart settings
68 of 68 selected

full table below

#ModelEffortScoreCost / taskHarness
1Fable 5.1max49.2%$13.18–
2Opus 5.5max45.6%$15.09–
3Opus 5max45.5%$5.14–
4Gemini 4high44.6%$3.47–
5Wonder Fullmax43.0%$2.15–
6Muse Spark 1.3max42.4%$0.390–
7Sonnet 5.5max42.0%$9.25–
8GLM 5.3max39.7%$1.80–
9Fable 5max38.3%$6.70–
10Wondermax36.3%$1.88–
11Grok 4.7xhigh34.2%$1.80–
12Kimi K3max33.4%$2.79–
13Grok 4.6high33.1%$0.978–
14Muse Spark 1.2xhigh32.8%$0.269–
15Hy4–32.6%$0.585–
16MiMo V2.6 Prohigh32.4%$0.134–
17Gemini 3.8 Flashhigh32.1%$0.857–
18Qwen 3.8 Maxmax32.1%$1.55–
19Haiku 5.5max31.9%$1.16–
20GLM 5.3 Flashmax31.6%$0.138–
21GPT-5.6 Solmax31.1%$7.51–
22Ember-1–31.1%$2.97–
23Qwen 3.8 27Bxhigh30.9%$0.815–
24Mistral Large 4high30.4%$2.72–
25GPT-5.6 Terramax30.0%$4.83–
26Opus 4.8max30.0%$1.69–
27Sonnet 5max29.4%$1.79–
28MiMo V2.6 Flashhigh28.5%$0.049–
29DeepSeek V4 Flash 0731high28.1%$0.156–
30Step 5 Preview–27.3%$0.553–
31DeepSeek V4 Pro 0813max26.8%$0.729–
32DeepSeek V4.1 Flashmax26.6%$0.135–
33Sonnet 4.6max25.8%$1.19–
34DeepSeek V4max25.2%$0.719–
35Grok 4.5high24.9%$0.474–
36GPT-5.6 Lunamax24.8%$0.465–
37Opus 4.7max23.7%$1.27–
38Gemini 3.7 Flashhigh22.8%$0.484–
39GPT-5.5xhigh21.8%$4.07–
40GPT-6.1 Solmax21.8%$1.32–
41Gemini 3.5 Flashhigh21.7%$0.860–
42GPT-6 Astramax20.7%$5.80–
43Ling 3.0 Flash Fin–20.6%$0.025–
44MiniMax M3max20.6%$0.164–
45GPT-6 Lunamax19.9%$0.191–
46Gemini 3.6 Flashhigh18.0%$0.356–
47Qwen 3.7 Maxmax15.8%$0.435–
48Inkling Smallxhigh15.4%$0.092–
49Inklingmax15.3%$0.354–
50GPT-6 Solmax15.0%$2.28–
51Kimi K2.6max12.4%$0.566–
52Gemini 3 Flashhigh11.8%$0.101–
53MiMo V2.5 Proxhigh11.5%$0.040–
54Qwen 3.7 Plushigh10.5%$0.094–
55Gemini 3.1 Prohigh9.3%$0.302–
56GPT-5.4 Minixhigh8.8%$0.806–
57MiMo V2.5xhigh7.8%$0.017–
58Qwen 3.6 Plusmax7.7%$0.298–
59Gemini 3.5 Flash-Litehigh7.3%$0.082–
60Grok 4.20xhigh7.0%$0.119–
61MiniMax M2.7high6.9%$0.120–
62Grok 4.3high6.2%$0.117–
63Haiku 4.5xhigh5.9%$0.226–
64GPT-5.4 Nanohigh5.0%$0.070–
65Mistral Medium 3.5high4.5%$1.44–
66Gemini 3.1 Flash-Litehigh3.1%$0.014–
67Nemotron 3.5 Lightning–2.6%$0.031–
68Mercury 2.5high2.5%$0.024–