Provider head-to-heads.

September 2026 prices · read daily · 10 pairings

Two providers at a time, on every model they both sell. Pairings are chosen rather than generated: ninety-six provider pages would make four thousand pairs, and most of them would have nothing in common to compare.

10 pairings, covering 505 side-by-side prices in all, each one the cheapest the provider charges under the standard blend of three parts input to one part output per million tokens.

• DeepInfravsFireworks AICheap hosting against hosting tuned for throughput.31 models both sell · DeepInfra cheaper on 27
• DeepInfravsGroqThe cheapest way to run open weights, or the fastest.10 models both sell · DeepInfra cheaper on 10
• DeepInfravsNebius Token FactoryA US host against a European one, on the same open models.26 models both sell · DeepInfra cheaper on 21
• DeepInfravsNovitaTwo hosts running the same open weights, competing on price alone.64 models both sell · DeepInfra cheaper on 48
• DeepInfravsOpenRouterBuying the open weights straight from a host, or through a router that shops around.97 models both sell · DeepInfra cheaper on 26
• DeepInfravsTogether AIThe price leader against the host with the longest catalog of the two.38 models both sell · DeepInfra cheaper on 36
• Fireworks AIvsTogether AIThe two hosts most often shortlisted together for production traffic.23 models both sell · Together AI cheaper on 7
• Nebius Token FactoryvsNovitaTwo regional hosts, and what running in Europe costs against running in Asia.21 models both sell · Novita cheaper on 11
• OpenRoutervsTogether AIOne key for everything, or a direct account with a host that runs the weights.45 models both sell · OpenRouter cheaper on 35
• OpenRoutervsVercel AI GatewayThe two routers most people reach for first, one API key over many makers.150 models both sell · OpenRouter cheaper on 37

cheapest price on any of these pages: $0.08 per million tokens