Qwen: Qwen2.5 VL 72B Instruct
qwen/qwen2.5-vl-72b-instruct
Cheapest provider
$0.25 / 1M
Nebius
Fastest provider (p95)
26 tok/s
Parasail
Intelligence (estimated)
71.0
Family + generation + popularity + price tier
Per-provider performance
Latency / throughput / uptime measured across providers over the last 30 minutes of live traffic. Atlas’s router weighs these per call (with the eval-gate signal) when picking a variant for Standard and Batch tiers.
| Provider | Quant | Input $/1M | Output $/1M | Latency p50 / p95 | Throughput p50 / p95 | Uptime 30m | Success |
|---|---|---|---|---|---|---|---|
| Nebius | q8· fp8 | $0.2500 | $0.7500 | 1272ms / 2874ms | 14 / 21 tok/s | 99.73% | 100.0% |
| Novita | full· bf16 | $0.8000 | $0.8000 | 3488ms / 6071ms | 10 / 23.3 tok/s | — | 65.7% |
| Parasail | q8· fp8 | $0.8000 | $1.0000 | 1429ms / 3337ms | 14 / 26 tok/s | 99.92% | 96.2% |
“—” 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 estimate
No public benchmark numbers indexed for this model yet, so the leaderboard score is derived from the catalogue data we sync: model family standing, generation, and live usage rank — applied identically to open- and closed-weight models. The heuristic ceiling sits below confirmed-benchmark frontier models so curated rankings stay clearly on top.
- ·qwen lg family base
Want a curated score? File a benchmark report and we’ll add it to the next sync.
How Atlas routes Qwen: Qwen2.5 VL 72B Instruct
- Realtime tier — direct passthrough at the upstream’s native precision. Best for hard latency / quality guarantees.
- Standard tier — Atlas picks the cheapest provider variant whose quantization has stayed green on your operation’s eval gates. For Qwen: Qwen2.5 VL 72B Instruct that’s currently Nebius at $0.25/1M.
- Batch tier — async, biggest discount. Roadmapped to use provider batch APIs (OpenAI / Anthropic) where available and queued spot capacity for open-weight workloads.