50 models7 providersprices scraped nightly — no estimates
mistralai · mixtral familyVerified May 18, 2026

Mixtral 8x7B Instruct

47B params · 33K context · apache-2.0 · released 2023-12 · 3 providers

Mixtral 8x7B Instruct is a 46.7-billion-parameter mixture-of-experts model from Mistral AI, released December 2023, activating roughly 13B parameters per token across 2 of 8 experts. It was among the first open-weights models to reach GPT-3.5-class quality on standard benchmarks, establishing MoE as a viable architecture for cost-efficient hosted inference. The 32K context window is adequate for most retrieval-augmented generation workloads, though tight by current standards. As of mid-2026 it remains broadly hosted — Fireworks, DeepInfra, Groq — but Mixtral 8x22B and Mistral Small 3 have largely displaced it for new deployments on both quality and cost curves. Apache 2.0 license; no commercial restrictions.

Cheapest right now
Fireworks AI$0.20/1M out
sticky while scrolling · verified May 18, 2026
Cheapest now
$0.20
Fireworks AI · fp16 · $/1M out
90-day price Δ
cheapest-host trend
Fastest TTFT
75 ms
Groq · 550 tok/s
Providers
3
indexed providers
✓ verified May 18, 2026
#Provider$/1M in$/1M outTTFT p50tok/sQuant
1Fireworks AICHEAPEST$0.20$0.20320 ms280fp16go ↗
2GroqFASTEST$0.24$0.2475 ms550fp8go ↗
3Hyperbolic$0.50$0.50450 ms90fp16go ↗

Prices verified May 18, 2026 by scraping 3 providers. Methodology · Raw data