Skip to main content
Bench

Language

English日本語中文Español

Site info

Task results

Research-level math problems (FrontierMath Tier 4 v2)

The FrontierMath expansion set of research-level problems, harder than the base set (Tiers 1-⁠3).

How to read this table

  • Higher rows are rated higher. A range like “1–2” means the gap is within the margin of error.
  • The score is the share of a shared set of problems the model solved (%).
  • A gap of 3 means 3 problems out of 100, and ± is how far the score moves when re-measured.
  • Prices are a guide per million tokens, reading then writing.
  • Length is how much text the model can read at once (tokens).
RankModelDeveloperAccuracy
(%)Accuracy scale 0 to 100		EvaluatedInput $
/1M		Output $
/1M		ContextSupportsWeightsReleasedPrice and specs
1–4GPT-6.1 SolOpenAIRating 100.0 ±0.0$2$101.05MOpenAI Released 2026-⁠09 Price $2 / $10 Length 1.05M
1–6GPT-6 AstraOpenAIRating 97.6 ±4.7$10$501.05MOpenAI Released 2026-⁠09 Price $10 / $50 Length 1.05M
1–9Claude Opus 5.5AnthropicRating 95.0 ±8.1$4$201MAnthropic Released 2026-⁠09 Price $4 / $20 Length 1M
2–12Claude Fable 5AnthropicRating 90.2 ±9.0$10$501MAnthropic Released 2026-⁠06 Price $10 / $50 Length 1M
1–13GPT-6 SolOpenAIRating 90.0 ±10.1$2$101.05MOpenAI Released 2026-⁠09 Price $2 / $10 Length 1.05M
2–16Claude Fable 5.1AnthropicRating 87.8 ±10.1$10$501MAnthropic Released 2026-⁠09 Price $10 / $50 Length 1M
3–16GPT-5.6 SolOpenAIRating 82.9 ±11.7$4$201.05MOpenAI Released 2026-⁠07 Price $4 / $20 Length 1.05M
3–16Claude Sonnet 5.5AnthropicRating 80.5 ±12.3$2$101MAnthropic Released 2026-⁠09 Price $2 / $10 Length 1M
3–16GPT-5.6 Sol ProOpenAIRating 80.5 ±12.3———OpenAI Released 2026-⁠07 Price — Length —
4–18Gemini 4 ArgonGoogleRating 78.0 ±12.8$2$101.05MGoogle Released 2026-⁠09 Price $2 / $10 Length 1.05M
4–18GPT-5.5 ProOpenAIRating 78.0 ±12.8$30$1801.05MOpenAI Released 2026-⁠04 Price $30 / $180 Length 1.05M
4–20AI co-mathematicianGoogleRating 75.6 ±13.1———Google Released 2026-⁠05 Price — Length —
5–20Claude Opus 5AnthropicRating 73.2 ±13.7$5$251MAnthropic Released 2026-⁠07 Price $5 / $25 Length 1M
6–20GPT-5.5OpenAIRating 72.5 ±14.0$5$301.05MOpenAI Released 2026-⁠04 Price $5 / $30 Length 1.05M
6–20GPT-5.6 TerraOpenAIRating 70.7 ±14.1$2$121.05MOpenAI Released 2026-⁠07 Price $2 / $12 Length 1.05M
6–20Muse WonderMetaRating 70.7 ±14.1$4$201.05MMeta Released 2026-⁠10 Price $4 / $20 Length 1.05M
10–26GPT-5.6 LunaOpenAIRating 61.0 ±15.1$0.2$1.21.05MOpenAI Released 2026-⁠07 Price $0.2 / $1.2 Length 1.05M
10–27GPT-5.4 ProOpenAIRating 58.5 ±15.3$30$1801.05MOpenAI Released 2026-⁠03 Price $30 / $180 Length 1.05M
12–28Claude Opus 4.8AnthropicRating 56.1 ±15.4$5$251MAnthropic Released 2026-⁠05 Price $5 / $25 Length 1M
12–28GPT-6 LunaOpenAIRating 56.1 ±15.4$0.1$0.51.05MOpenAI Released 2026-⁠09 Price $0.1 / $0.5 Length 1.05M
17–37GPT-5.4OpenAIRating 49.0 ±15.7$2.5$151.05MOpenAI Released 2026-⁠03 Price $2.5 / $15 Length 1.05M
17–44Claude Haiku 5.5AnthropicRating 46.3 ±15.5$0.1$0.51MAnthropic Released 2026-⁠10 Price $0.1 / $0.5 Length 1M
17–44DeepSeek-V4.1 Flash (2026-⁠10)⁠DeepSeekRating 46.3 ±15.5$0.15$0.61MDeepSeek Released 2026-⁠10 Price $0.15 / $0.6 Length 1M
17–44Muse Spark 1.3MetaRating 46.3 ±15.5$1.25$4.251.05MMeta Released 2026-⁠09 Price $1.25 / $4.25 Length 1.05M
17–44Qwen3.8 MaxAlibabaRating 46.3 ±15.5$2$61MAlibaba Released 2026-⁠08 Price $2 / $6 Length 1M
17–44GPT-5.2 ProOpenAIRating 46.0 ±15.3$21$168400KOpenAI Released 2025-⁠12 Price $21 / $168 Length 400K
18–49Kimi K3⁠MoonshotRating 39.0 ±15.1$3$151.05MMoonshot Released 2026-⁠07 Price $3 / $15 Length 1.05M
19–50Gemini 3.7 FlashGoogleRating 36.6 ±14.9$0.75$3.751.05MGoogle Released 2026-⁠08 Price $0.75 / $3.75 Length 1.05M
21–55Qwen3.7 MaxAlibabaRating 34.1 ±14.7$2.5$7.51MAlibaba Released 2026-⁠05 Price $2.5 / $7.5 Length 1M
21–55Qwen3.8 Max (2026-⁠09)AlibabaRating 34.1 ±14.7——1MAlibaba Released 2026-⁠09 Price — Length 1M
21–56Claude Opus 4.7AnthropicRating 31.7 ±14.4$5$251MAnthropic Released 2026-⁠04 Price $5 / $25 Length 1M
21–56Grok 4.6xAIRating 31.7 ±14.4$2$6500KxAI Released 2026-⁠08 Price $2 / $6 Length 500K
21–56Muse Spark Open⁠MetaRating 31.7 ±14.4——1.04MMeta Released 2026-⁠10 Price — Length 1.04M
21–56GPT-5.2OpenAIRating 31.7 ±14.3$1.75$14400KOpenAI Released 2025-⁠12 Price $1.75 / $14 Length 400K
21–59Claude Sonnet 5AnthropicRating 29.3 ±14.1$2$101MAnthropic Released 2026-⁠06 Price $2 / $10 Length 1M
21–59GLM-5.2⁠Z.aiRating 29.3 ±14.1$1.4$4.41MZ.ai Released 2026-⁠06 Price $1.4 / $4.4 Length 1M
21–59GLM-5.3⁠Z.aiRating 29.3 ±14.1$1.4$4.41MZ.ai Released 2026-⁠08 Price $1.4 / $4.4 Length 1M
22–59Claude Opus 4.6AnthropicRating 26.8 ±13.7$5$251MAnthropic Released 2026-⁠02 Price $5 / $25 Length 1M
22–59DeepSeek-V4.1 Flash⁠DeepSeekRating 26.8 ±13.7——1MDeepSeek Released 2026-⁠09 Price — Length 1M
22–59DeepSeek-V4 Pro (2026-⁠08)⁠DeepSeekRating 26.8 ±13.7$0.66$1.981MDeepSeek Released 2026-⁠08 Price $0.66 / $1.98 Length 1M

Best accuracy over time

2025-01-31Release date2026-10-17
How to read this page
  • Models solve them while running Python and submit the answer as a function. The value is accuracy.
  • The accuracy is the share of problems solved. A gap of 1 point is 1 question out of 100.
  • In the Supports column the icons are, from left: can use tools / can read images / can answer in a fixed format (JSON); faded means not supported or unknown.
  • Open-weights models have their contents (the weights) published, so you can also run them on your own computer or server. Models without the open-weights mark are normally used through the developer's own services (apps or API); for some models the weights status is unknown.
  • Names such as ChatGPT next to a model are the consumer services its developer offers. Check on the service itself whether that model is available there.

Data: Arena, Epoch AI, Swallow LLM team (CC BY 4.0), Models.dev (MIT), updated 2026-⁠10-⁠24, data sources