# FrontierMath v2 Tier 4 (FrontierMath v2 (Tier 4)) > Epoch AI's corrected v2 Tier 4 expansion, a separate set of exceptionally difficult research-level mathematics problems evaluated with Python-enabled iterative reasoning. Canonical page: https://benchlm.ai/benchmarks/frontiermathv2tier4 - Category: [Mathematics](/math) - Last updated: September 10, 2026 ## About FrontierMath v2 (Tier 4) - Year: 2026 - Tasks: 43 private extreme-difficulty mathematics problems - Format: Python-enabled iterative mathematical problem solving - Difficulty: Research-level mathematics requiring hours or days of expert work - Paper: [FrontierMath Tier 4 v2 leaderboard](https://epoch.ai/benchmarks/frontiermath-tier-4-v2?view=graph&tab=leaderboard) The v2 private Tier 4 set contains 43 problems. Its smaller sample makes individual scores noisier than Tiers 1-3, so it is a separate ranking factor with lower weight rather than being averaged into the core score. FrontierMath v2 (Tier 4) is currently weighted in BenchLM's scoring formula. The Mathematics category carries 5% of the overall score, and FrontierMath v2 (Tier 4) contributes 10% of that category score. ## Leaderboard (45 models) | Rank | Model | Creator | Score | |------|-------|---------|-------| | 1 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | 97.600% | | 2 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | OpenAI | 83.000% | | 3 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | OpenAI | 68.300% | | 4 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | OpenAI | 58.500% | | 5 | [GPT-5.5 Pro](/models/gpt-5-5-pro) | OpenAI | 39.600% | | 6 | [GPT-5.4 Pro](/models/gpt-5-4-pro) | OpenAI | 37.500% | | 7 | [GPT-5.5](/models/gpt-5-5) | OpenAI | 35.400% | | 8 | [Claude Opus 4.8](/models/claude-opus-4-8) | Anthropic | 31.250% | | 9 | [GPT-5.4](/models/gpt-5-4) | OpenAI | 27.100% | | 10 | [Claude Opus 4.7](/models/claude-opus-4-7) | Anthropic | 22.917% | | 11 | [Claude Opus 4.6](/models/claude-opus-4-6) | Anthropic | 22.900% | | 12 | [GPT-5.2](/models/gpt-5-2) | OpenAI | 18.800% | | 13 | [Gemini 3 Pro](/models/gemini-3-pro) | Google | 18.750% | | 14 | [Gemini 3.1 Pro](/models/gemini-3-1-pro) | Google | 16.700% | | 15 | [Muse Spark](/models/muse-spark) | Meta | 14.600% | | 16 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Google | 14.583% | | 17 | [Kimi K2.6](/models/kimi-2-6) | Moonshot AI | 14.580% | | 18 | [GLM-5.1](/models/glm-5-1) | Z.AI | 12.500% | | 19 | [GPT-5.1](/models/gpt-5-1) | OpenAI | 12.500% | | 20 | [Qwen3.6 Plus](/models/qwen3-6-plus) | Alibaba | 8.333% | | 21 | [Claude Sonnet 4.6](/models/claude-sonnet-4-6) | Anthropic | 8.300% | | 22 | [GPT-5.4 nano](/models/gpt-5-4-nano) | OpenAI | 6.250% | | 23 | [o4-mini (high)](/models/o4-mini-high) | OpenAI | 6.250% | | 24 | [Kimi K2.5](/models/kimi-k2-5) | Moonshot AI | 4.200% | | 25 | [Qwen 3.6 Max (preview)](/models/qwen3-6-max-preview) | Alibaba | 4.167% | | 26 | [Gemini 3 Flash](/models/gemini-3-flash) | Google | 4.167% | | 27 | [Claude Opus 4.5](/models/claude-opus-4-5) | Anthropic | 4.167% | | 28 | [Gemini 2.5 Pro](/models/gemini-2-5-pro) | Google | 4.167% | | 29 | [Claude Sonnet 4.5](/models/claude-sonnet-4-5) | Anthropic | 4.167% | | 30 | [Gemini 2.5 Flash](/models/gemini-2-5-flash) | Google | 4.167% | | 31 | [GLM-4.6](/models/glm-4-6) | Z.AI | 2.128% | | 32 | [GLM-5](/models/glm-5) | Z.AI | 2.100% | | 33 | [DeepSeek V3.2](/models/deepseek-v3-2) | DeepSeek | 2.100% | | 34 | [Grok 4](/models/grok-4) | xAI | 2.083% | | 35 | [Claude Haiku 4.5](/models/claude-haiku-4-5) | Anthropic | 2.083% | | 36 | [Qwen3.5 Plus](/models/qwen3-5-plus) | Alibaba | 2.083% | | 37 | [o3](/models/o3) | OpenAI | 2.083% | | 38 | [GPT-5.4 mini](/models/gpt-5-4-mini) | OpenAI | 2.080% | | 39 | [GLM-4.7](/models/glm-4-7) | Z.AI | 0.000% | | 40 | [Qwen3.5 Flash](/models/qwen3-5-flash) | Alibaba | 0.000% | | 41 | [Qwen3 235B 2507 (Reasoning)](/models/qwen3-235b-2507-reasoning) | Alibaba | 0.000% | | 42 | [Grok 3 [Beta]](/models/grok-3-beta) | xAI | 0.000% | | 43 | [GPT-4.1](/models/gpt-4-1) | OpenAI | 0.000% | | 44 | [Claude 3.5 Sonnet](/models/claude-3-5-sonnet) | Anthropic | 0.000% | | 45 | [Kimi K2](/models/kimi-k2) | Moonshot AI | 0.000% | ## FAQ ### What does FrontierMath v2 (Tier 4) measure? Epoch AI's corrected v2 Tier 4 expansion, a separate set of exceptionally difficult research-level mathematics problems evaluated with Python-enabled iterative reasoning. ### Which model scores highest on FrontierMath v2 (Tier 4)? GPT-6 Astra by OpenAI currently leads with a score of 97.600% on FrontierMath v2 (Tier 4). ### How many models are evaluated on FrontierMath v2 (Tier 4)? 45 AI models have been evaluated on FrontierMath v2 (Tier 4) on BenchLM. ### Does FrontierMath v2 (Tier 4) affect BenchLM's overall score? Yes. FrontierMath v2 (Tier 4) is a weighted benchmark inside the Mathematics category, which carries 5% of BenchLM's overall score. FrontierMath v2 (Tier 4) itself contributes 10% of that category score. ## Compare Top Models on FrontierMath v2 (Tier 4) - [GPT-6 Astra vs GPT-5.6 Sol](/compare/gpt-5-6-sol-vs-gpt-6-astra) - [GPT-5.6 Sol vs GPT-5.6 Terra](/compare/gpt-5-6-sol-vs-gpt-5-6-terra) - [GPT-5.6 Terra vs GPT-5.6 Luna](/compare/gpt-5-6-luna-vs-gpt-5-6-terra) - [GPT-5.6 Luna vs GPT-5.5 Pro](/compare/gpt-5-5-pro-vs-gpt-5-6-luna)