Best cell per row highlighted. Null means undisclosed — never counted as zero.
| Provider | Meta | OpenAI | |
| Family / tier | Llama 4 · Scout | Gemini 3 · Pro | GPT-5 · Pro |
| Status | GA | preview | GA |
| Released | 2025-04-05 | 2026-02-19 | 2026-04-24 |
| Knowledge cutoff | 2024-08 | 2025-01 | 2025-12 |
| Context window | 10M | 1.0M | 1.1M |
| Max output tokens | 8K | 66K | 128K |
| Input modalities | text, image | text, image, audio, video | text, image |
| Reasoning mode | none | optional | optional |
| Open weights | Yes | No | No |
| Input | $0.1 | $2 | $5 |
| Output | $0.34 | $12 | $30 |
| Cached input | — | $0.2 | $0.5 |
| Batch input | — | $1 | $2.5 |
| Free tier | No | No | No |
| Output speed (tok/s) | 106.1 | 142.7 | 68.7 |
| Time to first token (s) | 0.56 | 30.78 | 64 |
| Latency tier | fast | slow | slow |
| Trains on inputs | No | No | No |
| License | Llama-4-Community | Proprietary | Proprietary |
| Clouds | Bedrock, Vertex AI, Azure AI Foundry, GCP, OCI, IBM watsonx | Vertex AI, GCP | Azure OpenAI, Azure AI Foundry |
| Compliance | — | SOC2, HIPAA, GDPR, ISO27001, FedRAMP, CCPA | SOC2, GDPR, CCPA, HIPAA |
| Unified score | 7.5 | 8.7 | 8.7 |
| Decision Maker | 8.5 | 9 | 9 |
| Domain Strategist | 7.5 | 9 | 8.8 |
| Finance Lead | 9 | 8.5 | 7 |
| Domain Practitioner | 7.5 | 8.5 | 9.2 |
| Power User | 6.5 | 8.5 | 9 |
| Skeptic | 5.5 | 7.5 | 7.8 |
| Value score | 9.5 | 8.5 | 6.5 |
| MMLU | 79.6 | 92.6 | — |
| MMLU-Pro | 74.3 | — | 92.4 |
| GPQA Diamond | 57.2 | 94.3 | 93.6 |
| MATH-500 | 50.3 | — | — |
| SWE-bench Verified | — | 80.6 | 88.7 |
| HumanEval | 82 | — | — |
| LiveCodeBench | 32.8 | 2887 | — |
| MMMU | 69.4 | 80.5 | 81.2 |
| TAU-bench | — | 90.8 | 98 |
| Terminal-Bench | — | 68.5 | 82.7 |
| MRCR Long Context | — | 84.9 | 74 |
| Humanity's Last Exam | — | 44.4 | 41.4 |
| LMArena Elo | 1321 | 1487 | 1482 |
| LMArena Coding Elo | 1362 | 1520 | 1520 |
Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.