Benchmarks
Compare AI models
Pick two models to see release dates, benchmark scores, and community ratings side by side.
Frontier benchmarks
Frontier board
6 releases. Five ways to compare.
Selected for recent releases and benchmark coverage, not a single overall rank. Each column sorts the same models by its own score. No reading means no bar.
○ Independent evaluation · ◆ Vendor or self-reported, labeled per reading.
Coding
BenchLM · 0 to 100
- Gemini 4 Neon92.9Independent
Google DeepMind
- GPT-6.2 Astra91.9Independent
OpenAI
- Gemini 4 Carbon91.8Independent
Google DeepMind
- Claude Opus 5.588.5Independent
Anthropic
- GPT-6.2 Sol83.9Independent
OpenAI
- GPT-6 Astra82.2Independent
OpenAI
Agentic
BenchLM · 0 to 100
- GPT-6.2 Astra92.9Independent
OpenAI
- Gemini 4 Neon91.9Independent
Google DeepMind
- Gemini 4 Carbon88.6Independent
Google DeepMind
- GPT-6 Astra87.3Independent
OpenAI
- Claude Opus 5.587.1Independent
Anthropic
- GPT-6.2 Sol86.6Independent
OpenAI
Reasoning
BenchLM · 0 to 100
- Gemini 4 Neon88.9Independent
Google DeepMind
- Gemini 4 Carbon87.8Independent
Google DeepMind
- GPT-6.2 Astra87.3Independent
OpenAI
- GPT-6 Astra87Independent
OpenAI
- GPT-6.2 Sol81.5Independent
OpenAI
- Claude Opus 5.581Independent
Anthropic
Knowledge
BenchLM · 0 to 100
- GPT-6.2 Astra88.8Independent
OpenAI
- Gemini 4 Neon87.9Independent
Google DeepMind
- Gemini 4 Carbon83.4Independent
Google DeepMind
- GPT-6.2 Sol81.4Independent
OpenAI
- GPT-6 Astra80Independent
OpenAI
- Claude Opus 5.571.7Independent
Anthropic
Arena
LM Arena · Elo
- Claude Opus 5.51,508.73Independent
Anthropic
- GPT-6 Astra1,476.48Independent
OpenAI
- Gemini 4 CarbonNot reported
Google DeepMind
- Gemini 4 NeonNot reported
Google DeepMind
- GPT-6.2 AstraNot reported
OpenAI
- GPT-6.2 SolNot reported
OpenAI
Score links open the source; hover for capture details. Bar lengths start at zero. Arena lengths use the full public board maximum.
The shape of the scores
First three featured models. Toggle a name to isolate a line. Missing readings leave gaps, not zeroes.
3 of 3 series shown.
Axes share drawing space, not meaning. Arena is Elo divided by the current board maximum; it is not a percentage accuracy.
The newest arrival
Released 2026-11-20 · At least two readings
No predecessor is linked in the catalog yet. We do not infer a generation jump from a name.
Latest release with linked lineage
Claude Fable 5.1 vs Claude Fable 5
Latest snapshots, not a controlled before-and-after trial. Capture dates and evaluation setups may differ.
Scores and sources
| Model | Metric | Score / provenance | Captured | Source label / votes |
|---|---|---|---|---|
| Gemini 4 Neon | CodingBenchLM.ai | 92.9Independent | 2026-11-22 | Gemini 4 Neon |
| GPT-6.2 Astra | CodingBenchLM.ai | 91.9Independent | 2026-11-22 | GPT-6.2 Astra |
| Gemini 4 Carbon | CodingBenchLM.ai | 91.8Independent | 2026-11-22 | Gemini 4 Carbon |
| Claude Opus 5.5 | CodingBenchLM.ai | 88.5Independent | 2026-11-22 | Claude Opus 5.5 |
| GPT-6.2 Sol | CodingBenchLM.ai | 83.9Independent | 2026-11-22 | GPT-6.2 Sol |
| GPT-6 Astra | CodingBenchLM.ai | 82.2Independent | 2026-11-22 | GPT-6 Astra |
| GPT-6.2 Astra | AgenticBenchLM.ai | 92.9Independent | 2026-11-22 | GPT-6.2 Astra |
| Gemini 4 Neon | AgenticBenchLM.ai | 91.9Independent | 2026-11-22 | Gemini 4 Neon |
| Gemini 4 Carbon | AgenticBenchLM.ai | 88.6Independent | 2026-11-22 | Gemini 4 Carbon |
| GPT-6 Astra | AgenticBenchLM.ai | 87.3Independent | 2026-11-22 | GPT-6 Astra |
| Claude Opus 5.5 | AgenticBenchLM.ai | 87.1Independent | 2026-11-22 | Claude Opus 5.5 |
| GPT-6.2 Sol | AgenticBenchLM.ai | 86.6Independent | 2026-11-22 | GPT-6.2 Sol |
| Gemini 4 Neon | ReasoningBenchLM.ai | 88.9Independent | 2026-11-22 | Gemini 4 Neon |
| Gemini 4 Carbon | ReasoningBenchLM.ai | 87.8Independent | 2026-11-22 | Gemini 4 Carbon |
| GPT-6.2 Astra | ReasoningBenchLM.ai | 87.3Independent | 2026-11-22 | GPT-6.2 Astra |
| GPT-6 Astra | ReasoningBenchLM.ai | 87Independent | 2026-11-22 | GPT-6 Astra |
| GPT-6.2 Sol | ReasoningBenchLM.ai | 81.5Independent | 2026-11-22 | GPT-6.2 Sol |
| Claude Opus 5.5 | ReasoningBenchLM.ai | 81Independent | 2026-11-22 | Claude Opus 5.5 |
| GPT-6.2 Astra | KnowledgeBenchLM.ai | 88.8Independent | 2026-11-22 | GPT-6.2 Astra |
| Gemini 4 Neon | KnowledgeBenchLM.ai | 87.9Independent | 2026-11-22 | Gemini 4 Neon |
| Gemini 4 Carbon | KnowledgeBenchLM.ai | 83.4Independent | 2026-11-22 | Gemini 4 Carbon |
| GPT-6.2 Sol | KnowledgeBenchLM.ai | 81.4Independent | 2026-11-22 | GPT-6.2 Sol |
| GPT-6 Astra | KnowledgeBenchLM.ai | 80Independent | 2026-11-22 | GPT-6 Astra |
| Claude Opus 5.5 | KnowledgeBenchLM.ai | 71.7Independent | 2026-11-22 | Claude Opus 5.5 |
| Claude Opus 5.5 | ArenaLM Arena | 1,508.73Independent | 2026-10-26 | Claude Opus 5.5 (High)10,322 votes |
| GPT-6 Astra | ArenaLM Arena | 1,476.48Independent | 2026-10-26 | GPT-6 Astra (Max)14,076 votes |
| Gemini 4 Carbon | ArenaLM Arena | Not reported | Not reported | Not reported |
| Gemini 4 Neon | ArenaLM Arena | Not reported | Not reported | Not reported |
| GPT-6.2 Astra | ArenaLM Arena | Not reported | Not reported | Not reported |
| GPT-6.2 Sol | ArenaLM Arena | Not reported | Not reported | Not reported |
| Model | Metric / source | Score / provenance | Captured |
|---|---|---|---|
| Gemini 4 Neon | MathBenchLM.ai | 83.9Independent | 2026-11-22 |
| Gemini 4 Neon | MultimodalGroundedBenchLM.ai | 80.9Independent | 2026-11-22 |
| Gemini 4 Neon | InstructionFollowingBenchLM.ai | 86.9Independent | 2026-11-22 |
| GPT-6.2 Astra | MathBenchLM.ai | 83.5Independent | 2026-11-22 |
| GPT-6.2 Astra | InstructionFollowingBenchLM.ai | 83.3Independent | 2026-11-22 |
| GPT-6.2 Astra | MultimodalGroundedBenchLM.ai | 90.3Independent | 2026-11-22 |
| Gemini 4 Carbon | MathBenchLM.ai | 80.4Independent | 2026-11-22 |
| Gemini 4 Carbon | MultimodalGroundedBenchLM.ai | 85.9Independent | 2026-11-22 |
| Gemini 4 Carbon | InstructionFollowingBenchLM.ai | 82.6Independent | 2026-11-22 |
| GPT-6 Astra | Humanity's Last ExamEpoch AI | 54.8Independent | 2026-11-22 |
| GPT-6 Astra | MultimodalGroundedBenchLM.ai | 79.2Independent | 2026-11-22 |
| GPT-6 Astra | MathBenchLM.ai | 79.6Independent | 2026-11-22 |
| GPT-6 Astra | MultilingualBenchLM.ai | 74.1Independent | 2026-11-22 |
| GPT-6 Astra | InstructionFollowingBenchLM.ai | 78.6Independent | 2026-11-22 |
| Claude Opus 5.5 | MultimodalGroundedBenchLM.ai | 80.6Independent | 2026-11-22 |
| Claude Opus 5.5 | GDP.pdfEpoch AI | 48.8Independent | 2026-11-22 |
| Claude Opus 5.5 | MathBenchLM.ai | 77.5Independent | 2026-11-22 |
| Claude Opus 5.5 | MultilingualBenchLM.ai | 74.4Independent | 2026-11-22 |
| Claude Opus 5.5 | InstructionFollowingBenchLM.ai | 76.2Independent | 2026-11-22 |
| GPT-6.2 Sol | InstructionFollowingBenchLM.ai | 74.4Independent | 2026-11-22 |
| GPT-6.2 Sol | MultimodalGroundedBenchLM.ai | 84.2Independent | 2026-11-22 |
- Provider
- OpenAI-
- Out yet?
- Yes-
- Status
- Available-
- Release date
- Jul 9, 2026-
- LM Arena?LM Arena's Elo-style rating from blind head-to-head votes: people compare two anonymous model answers and pick the better one. Higher is better.
- 1485.1Verify at LM ArenaOct 26, 2026-
- Agentic?BenchLM.ai's score for multi-step agentic work - planning, tool use, and acting autonomously, normalized 0–100 across multiple benchmarks. Higher is better.
- 72.8Verify at BenchLM.aiNov 22, 2026-
- Coding?BenchLM.ai's score for code generation and software-engineering tasks, normalized 0–100 across multiple benchmarks. Higher is better.
- 74.2Verify at BenchLM.aiNov 22, 2026-
- InstructionFollowing?BenchLM.ai's score for following precise, detailed instructions, normalized 0–100 across multiple benchmarks. Higher is better.
- 67Verify at BenchLM.aiNov 22, 2026-
- Knowledge?BenchLM.ai's score for factual knowledge and question answering, normalized 0–100 across multiple benchmarks. Higher is better.
- 71.9Verify at BenchLM.aiNov 22, 2026-
- Math?BenchLM.ai's score for mathematical problem solving, normalized 0–100 across multiple benchmarks. Higher is better.
- 71.9Verify at BenchLM.aiNov 22, 2026-
- Multilingual?BenchLM.ai's score for capability across non-English languages, normalized 0–100 across multiple benchmarks. Higher is better.
- 68Verify at BenchLM.aiNov 22, 2026-
- MultimodalGrounded?BenchLM.ai's score for understanding grounded in images and documents, normalized 0–100 across multiple benchmarks. Higher is better.
- 71.7Verify at BenchLM.aiNov 22, 2026-
- Reasoning?BenchLM.ai's score for logic and multi-step reasoning problems, normalized 0–100 across multiple benchmarks. Higher is better.
- 73Verify at BenchLM.aiNov 22, 2026-
- GDP.pdf?Epoch AI's GDP.pdf benchmark: extracting and analyzing information from real-world PDF documents. Higher is better.Epoch AI
- 30.7Verify at Epoch AINov 17, 2026-
- HealthBench?OpenAI's HealthBench evaluation of healthcare conversation quality. This value is vendor-reported; higher is better.openai-system-card
- 57Verify at openai-system-cardJul 10, 2026-
- HealthBench Consensus?OpenAI's HealthBench Consensus evaluation. This value is vendor-reported; higher is better.openai-system-card
- 95.5Verify at openai-system-cardJul 10, 2026-
- HealthBench Hard?The harder subset of OpenAI's HealthBench healthcare evaluation. This value is vendor-reported; higher is better.openai-system-card
- 33.1Verify at openai-system-cardJul 10, 2026-
- HealthBench Professional?OpenAI's HealthBench Professional evaluation of model responses to realistic healthcare conversations. This value is vendor-reported; higher is better.openai-system-card
- 60.5Verify at openai-system-cardJul 10, 2026-
- Vibe rating?OutYet's community rating: signed-in users score the model 1–10. Shown as the average and the number of votes.
- 7.0 / 10 · 1-
- Predecessor
- --
- Successor
- --
| Open page → | ||
|---|---|---|
| Provider | OpenAI | - |
| Out yet? | Yes | - |
| Status | Available | - |
| Release date | Jul 9, 2026 | - |
| LM Arena?LM Arena's Elo-style rating from blind head-to-head votes: people compare two anonymous model answers and pick the better one. Higher is better. | 1485.1Verify at LM ArenaOct 26, 2026 | - |
| Agentic?BenchLM.ai's score for multi-step agentic work - planning, tool use, and acting autonomously, normalized 0–100 across multiple benchmarks. Higher is better. | 72.8Verify at BenchLM.aiNov 22, 2026 | - |
| Coding?BenchLM.ai's score for code generation and software-engineering tasks, normalized 0–100 across multiple benchmarks. Higher is better. | 74.2Verify at BenchLM.aiNov 22, 2026 | - |
| InstructionFollowing?BenchLM.ai's score for following precise, detailed instructions, normalized 0–100 across multiple benchmarks. Higher is better. | 67Verify at BenchLM.aiNov 22, 2026 | - |
| Knowledge?BenchLM.ai's score for factual knowledge and question answering, normalized 0–100 across multiple benchmarks. Higher is better. | 71.9Verify at BenchLM.aiNov 22, 2026 | - |
| Math?BenchLM.ai's score for mathematical problem solving, normalized 0–100 across multiple benchmarks. Higher is better. | 71.9Verify at BenchLM.aiNov 22, 2026 | - |
| Multilingual?BenchLM.ai's score for capability across non-English languages, normalized 0–100 across multiple benchmarks. Higher is better. | 68Verify at BenchLM.aiNov 22, 2026 | - |
| MultimodalGrounded?BenchLM.ai's score for understanding grounded in images and documents, normalized 0–100 across multiple benchmarks. Higher is better. | 71.7Verify at BenchLM.aiNov 22, 2026 | - |
| Reasoning?BenchLM.ai's score for logic and multi-step reasoning problems, normalized 0–100 across multiple benchmarks. Higher is better. | 73Verify at BenchLM.aiNov 22, 2026 | - |
| GDP.pdf?Epoch AI's GDP.pdf benchmark: extracting and analyzing information from real-world PDF documents. Higher is better.Epoch AI | 30.7Verify at Epoch AINov 17, 2026 | - |
| HealthBench?OpenAI's HealthBench evaluation of healthcare conversation quality. This value is vendor-reported; higher is better.openai-system-card | 57Verify at openai-system-cardJul 10, 2026 | - |
| HealthBench Consensus?OpenAI's HealthBench Consensus evaluation. This value is vendor-reported; higher is better.openai-system-card | 95.5Verify at openai-system-cardJul 10, 2026 | - |
| HealthBench Hard?The harder subset of OpenAI's HealthBench healthcare evaluation. This value is vendor-reported; higher is better.openai-system-card | 33.1Verify at openai-system-cardJul 10, 2026 | - |
| HealthBench Professional?OpenAI's HealthBench Professional evaluation of model responses to realistic healthcare conversations. This value is vendor-reported; higher is better.openai-system-card | 60.5Verify at openai-system-cardJul 10, 2026 | - |
| Vibe rating?OutYet's community rating: signed-in users score the model 1–10. Shown as the average and the number of votes. | 7.0 / 10 · 1 | - |
| Predecessor | - | - |
| Successor | - | - |
Benchmark scores are mirrored from third-party sources and captured on the dates shown. Numbers from different benchmarks, sources, or capture dates are not directly comparable.
Data from BenchLM.ai · Epoch AI, “AI Benchmarking Hub” (CC BY 4.0).