DeepSeek V3.x API prices across providers

Identical open weights, different invoices. All prices USD per 1M tokens, verified against provider sources. Last full sweep:

DeepSeek V3.x hosted inference prices by provider, USD per 1M tokens
Provider In $/M Out $/M Spread vs cheapest Context tok/s Last verified Go link
DeepInfra 0.27 1.1 1.00× cheapest 164k · manual-csv Go ↗

Cache-hit pricing excluded — model your real mix in the Cost Calculator (ships week 2). Spread = output price ÷ cheapest output price.

Measured floor: DeepInfra at $1.1/M out. Widest spread on this model today: 1.00×.

What a month costs

Worked examples at this model's verified prices (2026-08-23 sweep); cache-hit discounts excluded.

Monthly USD cost examples for DeepSeek V3.x by provider
Monthly workloadNovita AIDeepInfraTogetherOpenRouter
Light — 1M in / 0.25M out $0.59$0.55$0.72$0.69
Mid — 10M in / 2M out $5.26$4.9$6.6$6.18
Heavy — 50M in / 10M out $26.3$24.5$33$30.9

Head-to-heads: Qwen3 235B · DeepSeek R1 · Llama 4 Maverick

Also tracked: DeepSeek R1 · Qwen3 235B · Qwen3 Coder · Llama 4 Maverick · Kimi K2 · GLM 4.7 · GPT-OSS 120B · Gemma 3 27B · MiniMax M3

How we verify

  1. Pull — provider pricing pages and /models endpoints, daily at 06:00 UTC.
  2. Probe — fixed 1k-in/256-out prompt per provider; billed tokens and latency recorded.
  3. Diff — changes beyond ±1% raise a price move; every row carries its own last-verified date, never a site-wide stamp.
  4. Disclose — affiliate Go links are labeled and never reorder this table; methodology and raw probes are published.
Freshness policy: rows older than 14 days carry the amber stale flag until re-probed. Stale data beats fresh guesses.

Spot a wrong number? File a correction — we re-probe within one sweep cycle.