OpenRouter provider table for meta-llama/llama-3.3-70b-instruct, third re-read 2026-10-17 (per-provider price, P50 latency, throughput, uptime)
Fields
sources/svc-src-openrouter-provider-table-2026-10-17-r3Notes
Fetched later on 2026-10-17 than svc-src-openrouter-provider-table-2026-10-17-r2, same URL. Per-provider rows for Llama 3.3 70B Instruct, P50 over OpenRouter's own routed traffic. DeepInfra (Turbo) $0.100/$0.320, 0.55s latency, 16 tps, 99.94% uptime (was 0.46s / 26 tps / 99.93% on the r2 read). NovitaAI $0.135/$0.400, 1.08s, 41 tps, 99.25% (was 1.04s / 48 tps / 99.16%). AkashML $0.200/$0.520 (cache read $0.100), 0.62s, 38 tps, 99.85%. Parasail $0.220/$0.500 (cache read $0.110), 0.70s, 40 tps, 100.00%. Cloudflare $0.293/$2.253, 0.64s, 38 tps, 99.39%. SambaNova Turbo list $0.60/$1.20 struck through next to a 25% off badge rendering $0.450/$0.900, 0.78s, 110 tps, 99.88% (was 0.96s / 117 tps / 99.90%). Groq $0.590/$0.790 (cache read $0.295), 0.53s, 201 tps, 100.00% (was 0.49s / 209 tps). CoreWeave $0.710/$0.710 (cache read $0.710), 0.16s, 86 tps, 100.00% (was 0.13s / 85 tps). Google Vertex (US flag) $0.720/$0.720, 2.32s, 92 tps (was 1.24s / 89 tps); a second Vertex row at the same price has no latency cell. Together $1.040/$1.040, 0.71s, 39 tps, 96.30% (was 0.91s / 53 tps / 95.86%). Page headline price is still the floor route $0.10/$0.32. Best-across-providers throughput widget reads 201 tok/s (Groq); best latency widget reads 0.16s (CoreWeave). List prices on every row are unchanged from the r2 read, including the SambaNova struck-through discount. These are OpenRouter-measured P50 numbers over its own traffic, not vendor claims and not the Artificial Analysis long-prompt medians stored on the individual provider records. The Artificial Analysis Llama 3.3 70B provider page was re-fetched the same day: its rendered summary still says Groq 305.7 t/s, SambaNova 286.3 t/s, Vertex 153.6 t/s, DeepInfra 14.9 t/s, Azure 130.9 t/s, and Together AI Turbo 1.53s first token — the same medians already on the records (the comparison table still rounds SambaNova to 288, Vertex to 150, DeepInfra to 15, Together to 79 / 1.50, Azure to 132 / 2.28). No AA scalar was moved.
Cross-references
This record references no other records.