NEWMEN
← back to all models

Qwen: Qwen3 235B A22B Thinking 2507

qwen/qwen3-235b-a22b-thinking-2507

Alibaba (Qwen)open-weight262K context5 providersIntelligence 83.1

Cheapest provider

$0.10 / 1M

WandB

Fastest provider (p95)

99 tok/s

WandB

Intelligence (composite)

83.1

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.

ProviderQuantInput $/1MOutput $/1MLatency p50 / p95Throughput p50 / p95Uptime 30mSuccess
WandBfull· bf16$0.1000$0.1000395ms / 1017ms84 / 99 tok/s100.00%84.3%
Alibabaundisclosed$0.1495$1.4950638ms / 6083ms69 / 81 tok/s98.47%97.2%
DeepInfraq8· fp8$0.2300$2.3000615ms / 1742ms47 / 62.1 tok/s99.64%97.6%
AtlasCloudq8· fp8$0.2800$2.30001257ms / 3269ms24 / 38.2 tok/s97.96%98.0%
Novitaq8· fp8$0.3000$3.0000719ms / 3184ms49 / 63.9 tok/s81.3%

“—” 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

84.4

broad reasoning

Code

74.1

pass@1 (HumanEval / LiveCodeBench)

AIME 2025

92.3

math accuracy

GPQA Diamond

81.1

hard reasoning

Source: Qwen3-235B-A22B-Thinking-2507 model card (MMLU-Pro 84.4, GPQA 81.1, LiveCodeBench v6 74.1, AIME25 92.3); huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507

How Atlas mode sources Qwen: Qwen3 235B A22B Thinking 2507

  • Strict mode — pin Qwen: Qwen3 235B A22B Thinking 2507 exactly and we pass it straight through, sourced from the cheapest provider above. The same model, no substitutions — currently WandB 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.