Best cell per row highlighted. Null means undisclosed — never counted as zero.
| Provider | DeepSeek | Anthropic | Alibaba Cloud |
| Family / tier | DeepSeek V4 · Flash | Claude 4 · Haiku | Qwen3 · Large |
| Status | retired | GA | GA |
| Released | 2026-04-24 | 2025-10-15 | 2025-04-29 |
| Knowledge cutoff | 2026-02 | 2025-02 | 2024-10 |
| Context window | 1M | 200K | 131K |
| Max output tokens | 384K | 64K | 33K |
| Input modalities | text | text, image | text |
| Reasoning mode | optional | optional | optional |
| Open weights | Yes | No | Yes |
| Input | $0.14 | $1 | $0.2 |
| Output | $0.28 | $5 | $0.6 |
| Cached input | $0.0028 | $0.1 | — |
| Batch input | — | $0.5 | — |
| Free tier | Yes | Yes | Yes |
| Output speed (tok/s) | — | 101.8 | 68 |
| Time to first token (s) | — | 0.86 | 0.78 |
| Latency tier | fast | fast | medium |
| Trains on inputs | Yes | No | No |
| License | MIT | Proprietary | Apache-2.0 |
| Clouds | — | Bedrock, Vertex AI, Azure AI Foundry | GCP |
| Compliance | — | SOC2, HIPAA, GDPR, ISO27001 | — |
| Unified score | 8.7 | 8.4 | 8.5 |
| Decision Maker | 8.5 | 8.5 | 8 |
| Domain Strategist | 8.5 | 8.5 | 8.5 |
| Finance Lead | 9.7 | 10 | 9.5 |
| Domain Practitioner | 8.5 | 8.5 | 9 |
| Power User | 8 | 8 | 7.5 |
| Skeptic | 7.5 | 7.5 | 7 |
| Value score | 9.9 | 9.8 | 9.5 |
| MMLU-Pro | 86.2 | 76 | 82.8 |
| GPQA Diamond | 88.1 | 67.2 | 70 |
| AIME 2025 | — | — | 81.5 |
| SWE-bench Verified | 79 | 73.3 | — |
| LiveCodeBench | 91.6 | — | 70.7 |
| Terminal-Bench | — | 41.75 | — |
| MRCR Long Context | 78.7 | — | — |
| SimpleQA | 34.1 | — | — |
| Humanity's Last Exam | 34.8 | — | — |
| LMArena Elo | 1438 | 1415 | 1423 |
| LMArena Coding Elo | 1483 | 1482 | 1472 |
Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.