Vals TaxEval v2 (TaxEval v2)

We show this table for reference; we do not rank on it.

Data verified September 1, 2026

A Vals-created set of questions and responses to tax questions

Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.

TaxEval v2 score on TaxEval v2 — September 1, 2026

We mirror the published taxeval v2 score view for TaxEval v2. Muse Spark 1.2 leads the public snapshot at 80.38%, followed by Muse Spark 1.1 (79.72%) and Muse Spark (77.68%). We do not use these results to rank models overall.

1Closed

Muse Spark 1.2

TaxEval v2 score table (145 models)

TaxEval v2 score results for TaxEval v2
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	Muse Spark 1.2Metaxhigh reasoning	80.38%	Not reported	Closed
2	Muse Spark 1.1Metaxhigh reasoning	79.72%	Not reported	Closed
3	Muse SparkMeta	77.68%	Not reported	Closed
4	Claude Sonnet 4.6Anthropic	77.11%	Not reported	Closed
5	Claude Fable 5Anthropic	76.94%	Not reported	Closed
6	GPT-5.6 TerraOpenAIxhigh reasoning	76.17%	Not reported	Closed
7	GPT-5.6 LunaOpenAImax reasoning	76.17%	Not reported	Closed
8	Claude Opus 4.6 (Adaptive)Anthropic	75.96%	Not reported	Closed
9	Claude Fable 5.1Anthropic	75.96%	Not reported	Closed
10	Grok 3xAI	75.88%	Not reported	Unknown
11	GPT-5.2OpenAIxhigh reasoning	75.76%	Not reported	Closed
12	Kimi K3Moonshot AI	75.72%	Not reported	Open
13	Grok 4 Fast (Reasoning)xAI	75.70%	Not reported	Closed
14	Claude Opus 4.8Anthropic	75.63%	Not reported	Closed
15	Claude Sonnet 5Anthropic	75.63%	Not reported	Closed
16	GLM-5.3-FlashZ.AImax reasoning	75.59%	Not reported	Open
17	Qwen3.8 MaxAlibaba	75.55%	Not reported	Open
18	Inkling-SmallThinking Machines Lab0.99 reasoning	75.51%	Not reported	Open
19	Qwen3.7 MaxAlibaba	75.31%	Not reported	Closed
20	InklingThinking Machines Lab0.99 reasoning	75.31%	Not reported	Open
21	Claude Opus 4.7Anthropic	75.27%	Not reported	Closed
22	GPT-5 miniOpenAIhigh reasoning	75.22%	Not reported	Closed
23	Claude Opus 5Anthropic	75.14%	Not reported	Closed
24	GPT-4.1OpenAIhigh reasoning	75.06%	Not reported	Closed
25	GPT-5.5OpenAIxhigh reasoning	74.98%	Not reported	Closed
26	Gemini 3.6 FlashGooglehigh reasoning	74.86%	Not reported	Closed
27	GPT-5.1OpenAIhigh reasoning	74.86%	Not reported	Closed
28	Claude Opus 4.5 ThinkingAnthropic	74.86%	Not reported	Closed
29	o4-miniOpenAIhigh reasoning	74.78%	Not reported	Closed
30	GPT-5.6 SolOpenAImax reasoning	74.78%	Not reported	Closed
31	Gemini 3.7 FlashGooglehigh reasoning	74.73%	Not reported	Closed
32	Qwen3.6 PlusAlibaba	74.73%	Not reported	Closed
33	Kimi K2.6Moonshot AI	74.65%	Not reported	Open
34	o3OpenAIhigh reasoning	74.57%	Not reported	Closed
35	GPT-4oOpenAIhigh reasoning	74.53%	Not reported	Closed
36	Gemini 3.8 FlashGooglehigh reasoning	74.45%	Not reported	Closed
37	Gemini 3.5 FlashGooglehigh reasoning	74.37%	Not reported	Closed
38	Claude Opus 4.5Anthropic	74.33%	Not reported	Closed
39	o1OpenAIhigh reasoning	74.28%	Not reported	Closed
40	Kimi K2.5 ThinkingMoonshot AI	74.20%	Not reported	Unknown
41	Grok 4.20 0309 ReasoningxAI	74.12%	Not reported	Unknown
42	Claude 3.7 Sonnet ThinkingAnthropic	74.04%	Not reported	Closed
43	Qwen3 Max PreviewAlibaba	73.96%	Not reported	Unknown
44	GPT-5.4OpenAIxhigh reasoning	73.96%	Not reported	Closed
45	Gemini 3 Flash PreviewGooglehigh reasoning	73.88%	Not reported	Unknown
46	MiMo-V2.5-ProXiaomi	73.79%	Not reported	Closed
47	Claude Opus 4.1 20250805 ThinkingAnthropic	73.67%	Not reported	Unknown
48	Qwen3 MaxAlibaba	73.51%	Not reported	Closed
49	GPT-5OpenAIhigh reasoning	73.39%	Not reported	Unknown
50	GLM-5.2Z.AI	73.34%	Not reported	Open
51	Claude Sonnet 4.5 ThinkingAnthropic	73.30%	Not reported	Closed
52	Grok 4.1 Fast (Reasoning)xAI	73.14%	Not reported	Closed
53	Nemotron 3 Ultra 550b A55bNvidia	73.10%	Not reported	Unknown
54	Mistral Large 2512Mistral AI	73.06%	Not reported	Unknown
55	DeepSeek V4 Pro 0813DeepSeekmax reasoning	73.06%	Not reported	Open
56	Grok 3 Mini Fast Low ReasoningxAIlow reasoning	72.98%	Not reported	Unknown
57	Gemini 2.5 Pro Exp 03 25Google	72.89%	Not reported	Unknown
58	Gemini 3.1 Pro PreviewGooglehigh reasoning	72.88%	Not reported	Unknown
59	Gemini 2.5 Flash Preview 09 2025Google	72.73%	Not reported	Unknown
60	MiniMax M3MiniMax	72.73%	Not reported	Open
61	Gemini 3.5 Flash-LiteGooglehigh reasoning	72.61%	Not reported	Closed
62	Gemini 3 Pro PreviewGooglehigh reasoning	72.57%	Not reported	Unknown
63	Claude 3.7 SonnetAnthropic	72.40%	Not reported	Closed
64	Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle	72.40%	Not reported	Unknown
65	GLM-4.5Z.AI	72.40%	Not reported	Closed
66	GLM-5.3Z.AImax reasoning	72.36%	Not reported	Open
67	DeepSeek-R1DeepSeek	72.28%	Not reported	Open
68	Qwen3.5 FlashAlibaba	72.16%	Not reported	Closed
69	DeepSeek V4 Pro 0813DeepSeekmax reasoning	72.08%	Not reported	Open
70	Claude Sonnet 4 20250514 ThinkingAnthropic	72.00%	Not reported	Unknown
71	GPT-4.1 miniOpenAIhigh reasoning	71.91%	Not reported	Closed
72	Claude Opus 4Anthropic	71.91%	Not reported	Unknown
73	MiMo-V2.5Xiaomi	71.83%	Not reported	Closed
74	Gemini 3.1 Flash Lite PreviewGooglehigh reasoning	71.79%	Not reported	Unknown
75	Kimi K2 ThinkingMoonshot AI	71.71%	Not reported	Unknown
76	Grok 4.5xAIhigh reasoning	71.67%	Not reported	Closed
77	GPT-OSS 120BOpenAI	71.59%	Not reported	Open
78	Grok 4 Fast Non ReasoningxAI	71.58%	Not reported	Unknown
79	Claude Opus 4.1Anthropic	71.46%	Not reported	Unknown
80	Qwen3.6-27BAlibaba	71.26%	Not reported	Open
81	GPT-5.4 miniOpenAIxhigh reasoning	71.22%	Not reported	Closed
82	GLM-5.1Z.AI	71.19%	Not reported	Open
83	Gemini 2.5 Flash Preview 04 17Google	71.18%	Not reported	Unknown
84	GPT-4oOpenAIhigh reasoning	71.14%	Not reported	Closed
85	Grok 3 Mini Fast High ReasoningxAIhigh reasoning	71.14%	Not reported	Unknown
86	DeepSeek V3 0324DeepSeek	71.10%	Not reported	Open
87	Grok 4.6xAIhigh reasoning	71.10%	Not reported	Closed
88	Qwen3.8-27BAlibabaxhigh reasoning	70.85%	Not reported	Open
89	Grok 4.3xAI	70.81%	Not reported	Closed
90	DeepSeek V4 Flash 0731DeepSeekhigh reasoning	70.69%	Not reported	Open
91	Ling 3.0 Flash 2607Ant	70.65%	Not reported	Unknown
92	Qwen3 235B-A22BAlibaba	70.65%	Not reported	Open
93	Gemini 2.5 Flash Preview 04 17 ThinkingGoogle	70.52%	Not reported	Unknown
94	Mistral Medium 2505Mistral AI	70.32%	Not reported	Unknown
95	Moonshotai Kimi K2 InstructTogether AI	70.20%	Not reported	Unknown
96	Claude 3.5 SonnetAnthropic	70.16%	Not reported	Closed
97	GLM 5 ThinkingZhipu AI	70.03%	Not reported	Unknown
98	Gemini 2.0 Flash Thinking Exp 01 21Google	69.79%	Not reported	Unknown
99	Claude Sonnet 4Anthropic	69.62%	Not reported	Unknown
100	o3-miniOpenAIhigh reasoning	69.42%	Not reported	Closed
101	GLM-4.7Z.AI	68.77%	Not reported	Open
102	Gemini 2.0 Pro Exp 02 05Google	68.15%	Not reported	Unknown
103	MiniMax M2.5MiniMax	68.15%	Not reported	Closed
104	DeepSeek V3p2 ThinkingFireworks AIhigh reasoning	68.15%	Not reported	Unknown
105	Mistral Medium 3.5Mistral AIhigh reasoning	67.99%	Not reported	Unknown
106	DeepSeek V3DeepSeek	67.91%	Not reported	Open
107	Gemini 2.0 Flash ExpGoogle	67.74%	Not reported	Unknown
108	Claude Haiku 4.5 ThinkingAnthropic	67.54%	Not reported	Closed
109	GPT-5.4 nanoOpenAIhigh reasoning	67.42%	Not reported	Closed
110	GPT-5 nanoOpenAIhigh reasoning	67.38%	Not reported	Closed
111	Grok 2 1212xAI	67.05%	Not reported	Unknown
112	Llama4 Maverick Instruct BasicFireworks AI	66.56%	Not reported	Unknown
113	MiniMax M2.7MiniMax	66.56%	Not reported	Open
114	MiniMax M2.1MiniMax	66.35%	Not reported	Unknown
115	GLM-4.6Z.AI	66.23%	Not reported	Open
116	Gemini 2.5 Flash Lite Preview 09 2025Google	66.23%	Not reported	Unknown
117	Gemini 2.0 Flash 001Google	65.25%	Not reported	Unknown
118	Grok 4 0709xAI	65.09%	Not reported	Unknown
119	Gemini 2.5 Flash Lite Preview 09 2025 ThinkingGoogle	64.72%	Not reported	Unknown
120	Mistral Large 2411Mistral AI	63.78%	Not reported	Unknown
121	Langston Nim Nvidia Llama 3.3 Nemotron Super 49b V1 42e84561 ThinkingTogether AI	63.74%	Not reported	Unknown
122	GPT-OSS 20BOpenAI	63.70%	Not reported	Open
123	Magistral Medium 2509Mistral AI	61.94%	Not reported	Unknown
124	Grok 4.1 Fast Non ReasoningxAI	61.86%	Not reported	Unknown
125	Command A 03 2025Cohere	61.37%	Not reported	Unknown
126	Jamba Large 1.6AI21 Labs	60.88%	Not reported	Unknown
127	Meta Llama Meta Llama 3.1 405B Instruct TurboTogether AI	60.88%	Not reported	Unknown
128	GPT-4.1 nanoOpenAIhigh reasoning	60.75%	Not reported	Closed
129	GPT-4o miniOpenAIhigh reasoning	60.55%	Not reported	Closed
130	Magistral Small 2509Mistral AI	60.30%	Not reported	Unknown
131	Langston Nim Nvidia Llama 3.3 Nemotron Super 49b V1 42e84561Together AI	60.22%	Not reported	Unknown
132	Gemini 1.5 Pro 002Google	59.48%	Not reported	Unknown
133	Meta Llama Llama 3.3 70B Instruct TurboTogether AI	59.44%	Not reported	Unknown
134	Laguna XS.2Poolside	58.95%	Not reported	Open
135	Mistral Small 2503Mistral AI	58.30%	Not reported	Unknown
136	Jamba 1.5 LargeAI21 Labs	58.18%	Not reported	Unknown
137	Claude 3.5 HaikuAnthropic	57.36%	Not reported	Closed
138	Meta Llama Meta Llama 3.1 70B Instruct TurboTogether AI	56.17%	Not reported	Unknown
139	Meta Llama Llama 4 Scout 17B 16E InstructTogether AI	55.19%	Not reported	Unknown
140	Mistral Small 2402Mistral AI	49.14%	Not reported	Unknown
141	Gemini 1.5 Flash 002Google	48.20%	Not reported	Unknown
142	Jamba Mini 1.6AI21 Labs	44.60%	Not reported	Unknown
143	Jamba 1.5 MiniAI21 Labs	41.86%	Not reported	Unknown
144	Meta Llama Meta Llama 3.1 8B Instruct TurboTogether AI	32.34%	Not reported	Unknown
145	Laguna M.1Poolside	1.64%	Not reported	Closed

How TaxEval v2 is shown here

BenchLM mirrors the public Vals AI TaxEval v2 leaderboard captured from https://www.vals.ai/benchmarks/tax_eval_v2 and updated by Vals on September 1, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

TaxEval 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

145 Vals rows3 task viewsprivate datasetTasks: Overall, Correctness, Stepwise ReasoningDisplay only

TaxEval v2 on Vals AI Vals methodology Vals home

The published TaxEval v2 snapshot places Muse Spark 1.2 first at 80.38%. The third row is 2.70 points behind. The broader top-10 range is 4.50 points, so many of the published results sit in a relatively narrow band.

145 models have been evaluated on TaxEval v2. The benchmark falls in the Knowledge category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. TaxEval v2 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About TaxEval v2

Year

2026

Tasks

Tax question answering and response evaluation

Format

Accuracy score

Difficulty

Professional tax reasoning

BenchLM mirrors the public Vals AI TaxEval v2 leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

TaxEval v2 Public benchmark source

Freshness and provenance

Version

TaxEval v2 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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 TaxEval v2 measure?

A Vals-created set of questions and responses to tax questions

Which model leads the published TaxEval v2 snapshot?

Muse Spark 1.2 currently leads the published TaxEval v2 snapshot with 80.38% taxeval v2 score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on TaxEval v2?

The September 1, 2026 snapshot contains 145 AI models.

Last updated: September 1, 2026 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.