Best AI models for long arithmetic

Ranked by accuracy on the meo “long arithmetic” domain — one of eleven un-leaked, contamination-resistant domains. Per-domain strength reveals where a model genuinely reasons versus where it pattern-matches; the leaders differ sharply by domain.

Ranking 82 models across the meo-tested roster · as of 2026-10-18.

#	Model	Lab	accuracy
1	OpenAI: GPT-5.5	openai	100%
2	Anthropic: Claude Opus 5.5	anthropic	100%
3	OpenAI: GPT-5.5 Pro	openai	100%
4	OpenAI: GPT-6.1 Sol	openai	100%
5	Anthropic: Claude Fable 5	anthropic	100%
6	Anthropic: Claude Opus 4.8	anthropic	100%
7	MoonshotAI: Kimi K3	moonshotai	100%
8	Google: Gemini 4 Argon	google	93%
9	SpaceXAI: Grok 4.7	x-ai	93%
10	OpenAI: GPT-5.4 Nano	openai	93%
11	Nex AGI: Nex-N2-Pro (free)	nex-agi	73%
12	Z.ai: GLM 5.2	z-ai	60%
13	Google: Gemini 3.1 Pro Preview	google	53%
14	SpaceXAI: Grok 4.3	x-ai	47%
15	OpenAI: GPT-5 Nano	openai	47%
16	Qwen: Qwen3.5-Flash	qwen	40%
17	MoonshotAI: Kimi K2.7 Code	moonshotai	40%
18	OpenAI: GPT-5 Mini	openai	40%
19	inclusionAI: Ring-2.6-1T	inclusionai	33%
20	DeepSeek: DeepSeek V4 Flash 0423	deepseek	33%
21	Owl Alpha	openrouter	33%
22	Qwen: Qwen3.6 Flash	qwen	33%
23	MoonshotAI: Kimi K2.6	moonshotai	33%
24	OpenAI: GPT-5.1-Codex-Mini	openai	33%
25	Qwen: Qwen3 VL 30B A3B Instruct	qwen	33%
26	ByteDance Seed: Seed 1.6	bytedance-seed	33%
27	Qwen: Qwen3 VL 235B A22B Instruct	qwen	33%
28	Perceptron: Perceptron Mk1	perceptron	33%
29	ByteDance Seed: Seed-2.0-Lite	bytedance-seed	33%
30	Google: Gemini 3.5 Flash	google	27%
31	Qwen: Qwen3.7 Max	qwen	27%
32	Qwen: Qwen3 VL 30B A3B Thinking	qwen	27%
33	Qwen: Qwen3 VL 8B Thinking	qwen	27%
34	Mistral: Mistral Large 3 2512	mistralai	27%
35	Qwen: Qwen2.5 VL 72B Instruct	qwen	27%
36	Xiaomi: MiMo-V2.5-Pro	xiaomi	20%
37	Qwen: Qwen3.7 Plus	qwen	20%
38	Xiaomi: MiMo-V2.5	xiaomi	20%
39	Google: Gemma 4 31B	google	20%
40	OpenAI: GPT-4.1 Mini	openai	20%
41	Qwen: Qwen3 VL 8B Instruct	qwen	20%
42	OpenAI: GPT-4.1 Nano	openai	20%
43	DeepSeek: DeepSeek V4 Pro 0423	deepseek	13%
44	Qwen: Qwen3 VL 32B Instruct	qwen	13%
45	Z.ai: GLM 4.6V	z-ai	13%
46	Google: Gemini 3.1 Flash Lite Preview	google	13%
47	Qwen: Qwen3.5 Plus 2026-02-15	qwen	13%
48	Mistral: Ministral 3 14B 2512	mistralai	13%
49	StepFun: Step 3.7 Flash	stepfun	13%
50	NVIDIA: Nemotron 3 Nano Omni (free)	nvidia	13%
51	Amazon: Nova Lite 1.0	amazon	13%
52	NVIDIA: Nemotron 3 Ultra	nvidia	7%
53	MiniMax: MiniMax M3	minimax	7%
54	ByteDance Seed: Seed 1.6 Flash	bytedance-seed	7%
55	Mistral: Mistral Small 4	mistralai	7%
56	Google: Gemini 3.1 Flash Lite	google	7%
57	Tencent: Hy3 preview	tencent	7%
58	Mistral: Mistral Small 3.2 24B	mistralai	7%
59	MiniMax: MiniMax-01	minimax	7%
60	Anthropic: Claude 3 Haiku	anthropic	7%
61	Arcee AI: Trinity Large Thinking	arcee-ai	7%
62	Z.ai: GLM 5.1	z-ai	0%
63	Baidu: ERNIE 4.5 VL 424B A47B	baidu	0%
64	Google: Gemini 2.5 Flash Lite	google	0%
65	Google: Gemma 4 26B A4B	google	0%
66	Mistral: Ministral 3 8B 2512	mistralai	0%
67	Meta: Llama 4 Maverick	meta-llama	0%
68	Google: Gemma 3 27B	google	0%
69	Qwen: Qwen3.5-27B	qwen	0%
70	Google: Gemma 3 12B	google	0%
71	Meta: Llama 4 Scout	meta-llama	0%
72	OpenAI: GPT-4o-mini	openai	0%
73	Qwen: Qwen3.6 35B A3B	qwen	0%
74	Google: Gemma 3 4B	google	0%
75	Mistral: Ministral 3 3B 2512	mistralai	0%
76	Perplexity: Sonar	perplexity	0%
77	Qwen: Qwen3.6 Plus	qwen	0%
78	NVIDIA: Nemotron Nano 12B 2 VL (free)	nvidia	0%
79	Qwen: Qwen3.5-9B	qwen	0%
80	Qwen: Qwen3.5-35B-A3B	qwen	0%
81	ByteDance Seed: Seed-2.0-Mini	bytedance-seed	0%
82	Meta: Llama 3.2 11B Vision Instruct	meta-llama	0%

← All rankings Methodology & 𝕍 →