NEWMEN
← back to all models

Qwen: Qwen3 VL 8B Instruct

qwen/qwen3-vl-8b-instruct

Alibaba (Qwen)open-weight256K context4 providersIntelligence 58.0· est.

Cheapest provider

$0.08 / 1M

AtlasCloud

Fastest provider (p95)

114 tok/s

Alibaba

Intelligence (estimated)

58.0

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.

ProviderQuantInput $/1MOutput $/1MLatency p50 / p95Throughput p50 / p95Uptime 30mSuccess
AtlasCloudq8· fp8$0.0800$0.5000688ms / 2097ms34 / 49 tok/s99.42%97.4%
Novitaq8· fp8$0.0800$0.5000746ms / 1505ms35 / 52 tok/s99.68%93.7%
Alibabaundisclosed$0.1170$0.4550683ms / 1026ms14 / 114 tok/s100.00%100.0%
Parasailfull· bf16$0.2500$0.7500596ms / 2292ms35 / 98 tok/s100.00%93.6%

“—” 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.

Want a curated score? File a benchmark report and we’ll add it to the next sync.

How Atlas routes Qwen: Qwen3 VL 8B 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: Qwen3 VL 8B Instruct that’s currently AtlasCloud at $0.08/1M.
  • Batch tier — async, biggest discount. Roadmapped to use provider batch APIs (OpenAI / Anthropic) where available and queued spot capacity for open-weight workloads.