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

DeepSeek R1 Distill Llama 70B

70B params · 131K context · mit · released 2025-01 · 4 providers

DeepSeek R1 Distill Llama 70B is a 70-billion-parameter dense model from DeepSeek, released January 2025, produced by distilling reasoning-chain supervision from the full 671B R1 MoE into a Llama 3.3 70B base. Independent evals place it at roughly 70–80 percent of full R1's score on AIME and MATH benchmarks, at a fraction of the inference cost — the pragmatic pick when you need chain-of-thought reasoning but cannot justify 671B-MoE hosting overhead. Groq's hardware makes this one of the faster 70B options for latency-sensitive workloads; DeepInfra and Fireworks also carry it. MIT license means fully commercial-permissible deployment with no usage restrictions, which matters for enterprise teams with legal review requirements.

Cheapest right now
Hyperbolic$0.20/1M out
sticky while scrolling · verified May 27, 2026
Cheapest now
$0.20
Hyperbolic · fp16 · $/1M out
90-day price Δ
cheapest-host trend
Fastest TTFT
120 ms
Groq · 420 tok/s
Providers
4
indexed providers
✓ verified May 27, 2026
#Provider$/1M in$/1M outTTFT p50tok/sQuant
1HyperbolicCHEAPEST$0.20$0.20550 ms95fp16go ↗
2DeepInfra$0.28$0.55460 ms165fp16go ↗
3Fireworks AI$0.35$0.90420 ms180fp16go ↗
4DeepInfra$0.28$0.55460 ms165fp16go ↗
5GroqFASTEST$0.75$0.99120 ms420fp8go ↗

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