Best cell per row highlighted. Null means undisclosed — never counted as zero.
| Provider | Anthropic | Anthropic | Anthropic |
| Family / tier | Claude 4 · Sonnet | Claude 5 · Fable | Claude 4 · Opus |
| Status | GA | GA | GA |
| Released | 2026-02-16 | 2026-06-08 | 2026-05-27 |
| Knowledge cutoff | 2025-08 | — | — |
| Context window | 1M | 1M | 1M |
| Max output tokens | 64K | 128K | 128K |
| Input modalities | text, image | text, image, file | text, image |
| Reasoning mode | optional | always | always |
| Open weights | No | No | No |
| Input | $3 | $10 | $5 |
| Output | $15 | $50 | $25 |
| Cached input | $0.3 | $1 | $0.5 |
| Batch input | $1.5 | $5 | $2.5 |
| Free tier | Yes | No | No |
| Output speed (tok/s) | 75 | — | 60 |
| Time to first token (s) | 1.8 | — | 0.5 |
| Latency tier | fast | slow | medium |
| Trains on inputs | No | No | No |
| License | Proprietary | Proprietary | Proprietary |
| Clouds | Bedrock, Vertex AI, Azure AI Foundry | Bedrock, Vertex AI, Azure AI Foundry | Bedrock, Vertex AI, Azure AI Foundry |
| Compliance | SOC2, HIPAA, GDPR, ISO27001 | SOC2, HIPAA, GDPR, ISO27001 | SOC2, HIPAA, GDPR, ISO27001 |
| Unified score | 8.8 | 9.5 | 9.2 |
| Decision Maker | 9 | 9.5 | 9 |
| Domain Strategist | 9 | 9.5 | 9 |
| Finance Lead | 9 | 7 | 8.5 |
| Domain Practitioner | 9 | 9.5 | 9.5 |
| Power User | 8.5 | 9 | 9 |
| Skeptic | 8 | 7.5 | 8 |
| Value score | 9 | 6.5 | 8 |
| MMLU | 89.1 | — | — |
| MMLU-Pro | 87.3 | — | — |
| GPQA Diamond | 74.1 | — | 93.6 |
| AIME 2025 | 94 | — | — |
| MATH-500 | 89 | — | — |
| SWE-bench Verified | 79.6 | 95 | 88.6 |
| HumanEval | 98 | — | — |
| LiveCodeBench | 79.7 | — | — |
| MMMU | 83.6 | — | — |
| Terminal-Bench | — | 88 | 74.6 |
| LMArena Elo | 1460 | — | — |
| LMArena Coding Elo | 1500 | — | — |
| Artificial Analysis Index | 51 | — | — |
Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.