50 models7 providersprices scraped nightly — no estimates
alibaba · qwen familyVerified May 27, 2026

Qwen 2.5 72B Instruct

72B params · 131K context · qwen · released 2024-09 · 4 providers

Qwen 2.5 72B Instruct is the previous-generation flagship from Alibaba, a 72-billion-parameter model with a 131K context window released in September 2024. Superseded by Qwen 3 72B in April 2025, it remains widely deployed because workloads pinned to a specific checkpoint for reproducibility rarely migrate quickly — inference pipelines with cached prompts, fine-tuned adapters, or benchmarked output distributions commonly run a generation behind for six months or more. Benchmark scores on MMLU, HumanEval, and multilingual evals remain competitive against current-generation models in its parameter class. The Qwen license covers commercial use. Pricing across hosted providers tracks closely with Qwen 3 72B for now, though expect 2.5 rates to soften as provider capacity shifts toward the newer generation.

Cheapest right now
DeepInfra$0.35/1M out
sticky while scrolling · verified May 27, 2026
Cheapest now
$0.35
DeepInfra · fp16 · $/1M out
90-day price Δ
cheapest-host trend
Fastest TTFT
390 ms
Fireworks AI · 240 tok/s
Providers
4
indexed providers
✓ verified May 27, 2026
#Provider$/1M in$/1M outTTFT p50tok/sQuant
1DeepInfraCHEAPEST$0.18$0.35400 ms200fp16go ↗
2Fireworks AIFASTEST$0.20$0.80390 ms240fp16go ↗
3OpenRouter$0.22$0.75unknowngo ↗
4DeepInfra$0.18$0.35400 ms200fp16go ↗
5Hyperbolic$0.40$0.40500 ms105fp16go ↗

Prices verified May 27, 2026 by scraping 4 providers. Methodology · Raw data