MoonshotAI: Kimi K2 0905
moonshotai/kimi-k2-0905
Cheapest provider
$0.60 / 1M
AtlasCloud
Fastest provider (p95)
317 tok/s
Groq
Intelligence (composite)
69.5
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 |
|---|---|---|---|---|---|---|---|
| AtlasCloud | q8· fp8 | $0.6000 | $2.5000 | 856ms / 1508ms | 14 / 20 tok/s | 99.93% | 99.9% |
| Novita | q8· fp8 | $0.6000 | $2.5000 | 1740ms / 3421ms | 7 / 17 tok/s | 99.42% | 99.4% |
| Groq | undisclosed | $1.0000 | $3.0000 | 180ms / 581ms | 101 / 317.4 tok/s | 99.98% | 100.0% |
“—” 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
82.0
broad reasoning
Code
61.0
pass@1 (HumanEval / LiveCodeBench)
AIME 2025
57.0
math accuracy
GPQA Diamond
77.0
hard reasoning
Source: Artificial Analysis (AA-aligned) via MiniMax-M2 benchmark table — Kimi K2 0905 (MMLU-Pro 82, GPQA-D 77, LiveCodeBench 61, AIME25 57); github.com/MiniMax-AI/MiniMax-M2
How Atlas mode sources MoonshotAI: Kimi K2 0905
- Strict mode — pin MoonshotAI: Kimi K2 0905 exactly and we pass it straight through, sourced from the cheapest provider above. The same model, no substitutions — currently AtlasCloud at $0.60/1M.
- 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.