Offerings

Every active offering: a model as served by one host, with its current canonical price. Sort by clicking a column header; filter with the row below. Prices are USD per 1M tokens unless the unit toggle is set to per 1M characters (an approximation using a measured, blended tokens-per-character ratio).

Showing tier-tagged and vendor-direct models. Show all offerings, including the long tail

More columns: + Vendor + Region + Cache + Batch + Tokenizer - Last change + Source All columns Default columns

ModelHostVariantRegionInputOutputBlended ↑?ContextLast changeConfidence?
DeepSeek V4 FlashParasailfp8–0.1400.2800.1751,048,576–api
Gemini 2.5 Flash-Lite 3 sourcesGoogle Vertex AI––0.1000.4000.1751,048,576–api
Gemini 2.5 Flash-Lite 4 sourcesGoogle––0.1000.4000.1751,048,576–official
Llama 3.3 Nemotron Super 49B v1 2 sourcesDeepInfra––0.1000.4000.175131,072–aggregator
Llama 3.3 Nemotron Super 49B v1Nebius AI Studio––0.1000.4000.175131,072–aggregator
DeepSeek V4 FlashNebius AI Studio––0.1400.2800.1751,048,576–aggregator
DeepSeek V4 Flash 2 sourcesNovita AI––0.1400.2800.1751,048,576–aggregator
Llama 3.3 70B InstructNebius AI Studio––0.1300.4000.198128,000–aggregator
GPT-6 LunaOpenRouter––0.1000.5000.2001,050,000–api
GPT-6 Luna 4 sourcesOpenAI––0.1000.5000.2001,050,000–official
GPT-6 Luna 3 sourcesAzure AI Foundry––0.1000.5000.2001,050,000–api
GPT-6 Luna 2 sourcesAWS Bedrock––0.1000.5000.2001,050,000–aggregator
GPT-6 LunaPerplexity––0.1000.5000.2001,050,000–aggregator
GPT-6 Luna 2 sourcesAWS Bedrock–global0.1000.5000.2001,050,000–api
Claude Haiku 5.5 3 sourcesAWS Bedrock––0.1000.5000.2001,000,000–api
Claude Haiku 5.5 4 sourcesAnthropic––0.1000.5000.2001,000,000–official
Claude Haiku 5.5Azure AI Foundry–global0.1000.5000.2001,000,000–api
Claude Haiku 5.5Claude on AWS––0.1000.5000.2001,000,000–api
Claude Haiku 5.5Google Vertex AI–global0.1000.5000.2001,000,000–api
Claude Haiku 5.5OpenRouter––0.1000.5000.2001,000,000–api
Claude Haiku 5.5AWS Bedrock–global0.1000.5000.2001,000,000–aggregator
Claude Haiku 5.5Perplexity––0.1000.5000.2001,000,000–aggregator
Claude Haiku 5.5 2 sourcesAzure AI Foundry––0.1000.5000.2001,000,000–aggregator
Claude Haiku 5.5 2 sourcesGoogle Vertex AI––0.1000.5000.2001,000,000–aggregator
Llama 3.3 70B Instruct 3 sourcesNovita AI––0.1350.4000.201128,0002026-01-13 ↓api
Phi-4 2 sourcesAzure AI Foundry––0.1250.5000.21916,384–aggregator
GPT-6 Luna 2 sourcesAzure AI Foundry–us0.1100.5500.2201,050,000–api
GPT-6 Luna 2 sourcesAzure AI Foundry–eu0.1100.5500.2201,050,000–api
GPT-6 LunaAWS Bedrock–us-east-10.1100.5500.2201,050,000–api
GPT-6 LunaAWS Bedrock–us0.1100.5500.2201,050,000–aggregator
Claude Haiku 5.5AWS Bedrock–eu-west-10.1100.5500.2201,000,000–api
Claude Haiku 5.5AWS Bedrock–us-east-10.1100.5500.2201,000,000–api
Claude Haiku 5.5Azure AI Foundry–us0.1100.5500.2201,000,000–api
Claude Haiku 5.5Google Vertex AI–europe0.1100.5500.2201,000,000–api
Claude Haiku 5.5Google Vertex AI–us0.1100.5500.2201,000,000–api
Claude Haiku 5.5AWS Bedrock–apac0.1100.5500.2201,000,000–aggregator
Claude Haiku 5.5AWS Bedrock–au0.1100.5500.2201,000,000–aggregator
Claude Haiku 5.5AWS Bedrock–eu0.1100.5500.2201,000,000–aggregator
Claude Haiku 5.5AWS Bedrock–jp0.1100.5500.2201,000,000–aggregator
Claude Haiku 5.5AWS Bedrock–us0.1100.5500.2201,000,000–aggregator
Llama 4 ScoutDeepInfrafp8+priority–0.1500.4500.22510,000,000–api
Claude Haiku 5.5AWS Bedrock–us-gov-east-10.1200.6000.2401,000,000–aggregator
Claude Haiku 5.5AWS Bedrock–us-gov-west-10.1200.6000.2401,000,000–aggregator
Claude Haiku 5.5AWS Bedrock–us-gov0.1200.6000.2401,000,000–aggregator
DeepSeek V4 FlashMancerfp8–0.1750.5000.2561,048,5762026-11-07 ↓api
Mistral Small 4OpenRouter––0.1500.6000.263262,144–api
Mistral Small 4Mistral AIzdr–0.1500.6000.263262,144–api
Mistral Small 4 3 sourcesMistral AI––0.1500.6000.263262,1442026-08-21 ↑api
DeepSeek V4 Flash 2 sourcesAzure AI Foundry––0.1900.5100.2701,048,576–aggregator
Qwen3-235B-A22B 2 sourcesDeepInfra––0.1800.5400.270131,0722026-08-27 ↓aggregator

Confidence: official = vendor pricing page · api = OpenRouter per-host · aggregator = community list, may lag · unreviewed = automatic extraction awaiting review. When sources disagree the most authoritative wins. Details: /methodology#confidence