DeepSeek V4.1 Flash
The value champion: $0.26 per index task, with a LiveBench design score of 83.4 that matches Opus
Benchmark scores
How it is measured →Ranked 15 · bars are scaled to the field maximum for that benchmark.
| Benchmarks | Score | Vs. best in field | Gap |
|---|---|---|---|
| AA Intelligence Index v5 Intelligence index | 33.0 | 55% | −27.0 pts |
| AA Intelligence Index v4.3.2 Intelligence index (legacy) | 45.1 | 78% | −12.5 pts |
| GPQA Diamond Graduate-level science reasoning | 89.8 | 93% | −6.6 pts |
| Terminal-Bench 4.0 Terminal-agent tasks | 19.7 | 29% | −47.5 pts |
| LiveBench Design Document generation (DOCX/PPT/PDF) | 83.4 | 96% | −3.5 pts |
| Vals Index Vals composite index | 50.9 | 70% | −22.2 pts |
Assessment
A 552B backbone with 8B (prefill) / 16B (decode) active, 1M of context and an MIT licence with no commercial restrictions. AA Intelligence Index v5 33.0 — second among open models behind Kimi K3.1 Pro at 46.0; the earlier claim of being tied with GLM-5.3 for first no longer holds now that Kimi's weights are out. Its LiveBench design score of 83.4 is 3.5 points behind Opus 5.5 at 86.9, at 1/17 of the price. Cache hits cost $0.003 off-peak, the deepest cache discount in the industry. Mind the 2x peak/off-peak spread.
At the same price
The four models closest on blended price, side by side.
| Model | v5 | Blended | tok/s | Score gap |
|---|---|---|---|---|
| MiMo-V2.6 Pro Open weightsXiaomi | 30.0 | $0.61 | 81 | -3.0 |
| MiniMax M3.1 FlashMiniMax | 29.0 | $0.26 | 118 | -4.0 |
| Claude Haiku 5.5Anthropic | 36.0 | $0.20 | 235 | +3.0 |
| Mistral Large 4 Open weightsMistral | 26.0 | $1.05 | — | -7.0 |
Compare side by side
Line it up against other models on price, speed and every benchmark.
The comparison tool lines up to 4 models side by side with switchable normalisation — built for purchasing decisions.
Open the comparison tool → Back to leaderboard