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

Llama 3.1 70B Instruct

70B params · 131K context · llama-3 · released 2024-07 · 5 providers

Meta's Llama 3.1 70B Instruct is a 70-billion-parameter open-weights text model released in July 2024 under the Llama 3 community license. The 3.1 generation was the first in the 70B class to ship with a 131K-token context window, making it viable for long-document tasks like contract analysis and codebase summarization that previously required larger, costlier models. On MMLU it scores around 79–80, competitive with contemporaries at this parameter count. Note that Llama 3.3 70B, released December 2024, targets the same footprint with meaningfully better instruction-following — new deployments should default to 3.3 unless pinning to a specific weight hash for reproducibility. With continued provider adoption and price compression at the 70B tier, per-token costs here will keep falling through 2026.

Cheapest right now
DeepInfra$0.40/1M out
sticky while scrolling · verified May 27, 2026
Cheapest now
$0.40
DeepInfra · fp16 · $/1M out
90-day price Δ
cheapest-host trend
Fastest TTFT
88 ms
Groq · 575 tok/s
Providers
5
indexed providers
✓ verified May 27, 2026
#Provider$/1M in$/1M outTTFT p50tok/sQuant
1DeepInfraCHEAPEST$0.23$0.40430 ms195fp16go ↗
2Fireworks AI$0.22$0.88390 ms240fp16go ↗
3DeepInfra$0.23$0.40430 ms195fp16go ↗
4Hyperbolic$0.40$0.40510 ms105fp16go ↗
5Cerebras Inference$0.60$0.60210 ms2100fp16go ↗
6GroqFASTEST$0.59$0.7988 ms575fp8go ↗

Prices verified May 27, 2026 by scraping 5 providers. Methodology · Raw data