| Claude Fable 5 | Gemini 3 Ultra | |
|---|---|---|
| arenaElo | 1492 | 1441 |
| humanEval | 98 | 91 |
| sweBench | 77 | 66 |
| mmlu | 95.1 | 93.4 |
| gpqa | 92 | 72.6 |
| aime | 94 | 93 |
| Price in/out ($/M) | $10 / $50 | $18 / $72 |
Gemini 3 Ultra has the larger context window (3M vs 1M) and native video, which Fable 5 does not match. Fable 5 leads on raw capability (frontier reasoning, long-horizon agents and coding) at a lower list price ($10 / $50 vs $18 / $72). For whole-movie or 2000-page workloads on Vertex AI, Ultra. For the hardest agentic and reasoning work where your data fits in 1M tokens, Fable 5.