DeepSeek: DeepSeek V4 Flash
deepseek/deepseek-v4-flash
Cheapest provider
$0.10 / 1M
Baidu
Fastest provider (p95)
—
No throughput data yet — populated as traffic accumulates
Intelligence (composite)
88.5
MMLU-Pro · HumanEval · math · GPQA
Per-provider performance
Latency / throughput / uptime / price measured across providers over the last 30 minutes of live traffic. This is what proves “sourced cheapest” — Atlas mode draws on these per call to serve the cheapest path that holds quality.
| Provider | Quant | Input $/1M | Output $/1M | Latency p50 / p95 | Throughput p50 / p95 | Uptime 30m | Success |
|---|---|---|---|---|---|---|---|
| Baidu | q8· fp8 | $0.0983 | $0.1966 | — | — | 99.52% | — |
| DeepInfra | q4· fp4 | $0.1000 | $0.2000 | — | — | 97.98% | — |
| GMICloud | q8· fp8 | $0.1120 | $0.2240 | — | — | 96.48% | — |
| SiliconFlow | q8· fp8 | $0.1300 | $0.2800 | — | — | 95.45% | — |
| StreamLake | undisclosed | $0.1330 | $0.2660 | — | — | 99.14% | — |
| Alibaba | undisclosed | $0.1340 | $0.2680 | — | — | 99.86% | — |
| Morph | undisclosed | $0.1390 | $0.2780 | — | — | 92.30% | — |
| AkashML | q8· fp8 | $0.1400 | $0.2800 | — | — | 91.94% | — |
| AtlasCloud | q8· fp8 | $0.1400 | $0.2800 | — | — | 98.97% | — |
| DeepSeek | undisclosed | $0.1400 | $0.2800 | — | — | 99.98% | — |
| DigitalOcean | undisclosed | $0.1400 | $0.2800 | — | — | 99.49% | — |
| Novita | q8· fp8 | $0.1400 | $0.2800 | — | — | 99.36% | — |
| Parasail | q8· fp8 | $0.1400 | $0.2800 | — | — | 98.85% | — |
| Venice | undisclosed | $0.1700 | $0.3500 | — | — | 99.20% | — |
“—” means live telemetry hasn’t accumulated enough recent traffic for that endpoint. “undisclosed” means the provider serves the model but doesn’t expose the quantization label (typically running fp8 / int8 internally).
Intelligence breakdown
Composite score is a weighted average of public benchmarks (30% MMLU-Pro, 25% code pass@1, 25% math, 20% GPQA). Numbers come from model cards and the Artificial Analysis intelligence harness; missing components are renormalised over what’s present.
MMLU-Pro
86.2
broad reasoning
Code
91.6
pass@1 (HumanEval / LiveCodeBench)
MATH
—
math accuracy
GPQA Diamond
88.1
hard reasoning
Source: DeepSeek V4-Flash Max (MMLU-Pro 86.2, GPQA-Diamond 88.1, LiveCodeBench 91.6); framia benchmark table
How Atlas mode sources DeepSeek: DeepSeek V4 Flash
- Strict mode — pin DeepSeek: DeepSeek V4 Flash exactly and we pass it straight through, sourced from the cheapest provider above. The same model, no substitutions — currently Baidu at $0.10/1M.
- Atlas mode — the default. Each call is auto-optimized for the cheapest path that holds quality, at least 5% off going direct from call one and climbing as it ramps. You always see which model served the call and exactly what you saved — thumbs-down anything you don’t like for a full refund.