Model comparisons
Compare LLM API cost, model against model.
Each comparison prices the same prompt on two models at their current rates. Paste your own prompt on any page to see what it costs on both. Card rates are input and output prices per 1 million tokens.
- Comparisons
- 21
- Models priced
- 43
- Providers
- 9
- Rates verified
- 2026-10-02
The most capable models
Top-tier models, where output price and long-context tiers decide the bill.
- GPT-6 Astra vs Claude Fable 5.1Same headline price, four-fold cached-input gapGPT-6 Astra$10.00/$50.00Claude Fable 5.1$10.00/$50.00Open comparison
- Claude Sonnet 5.5 vs Claude Opus 5.5Half the token price, more cost per finished taskClaude Sonnet 5.5$2.00/$10.00Claude Opus 5.5$4.00/$20.00Open comparison
- Muse Spark 1.3 vs Claude Opus 5Three index points behind, at a quarter of the cost per taskMuse Spark 1.3$1.25/$4.25Claude Opus 5$5.00/$25.00Open comparison
Everyday production models
The models most teams run in production, priced between the frontier and the budget tier.
Cheap models for high volume
Models priced for millions of requests, where the monthly total matters more than one call.
- GPT-6 Luna vs GPT-5.6 LunaOpenAI halved the price of its cheapest tierGPT-6 Luna$0.10/$0.50GPT-5.6 Luna$0.20/$1.20Open comparison
- DeepSeek V4.1 Flash vs GPT-5.6 LunaHigh-volume pricing, off-peak and peakDeepSeek V4.1 Flash$0.15/$0.60GPT-5.6 Luna$0.20/$1.20Open comparison
- DeepSeek V4.1 Flash vs Qwen3.8 FlashThe budget price war, with peak-hour mathsDeepSeek V4.1 Flash$0.15/$0.60Qwen3.8 Flash$0.15/$0.47Open comparison
- Gemini 3.5 Flash-Lite vs GPT-5.6 LunaMultimodal image and document costGemini 3.5 Flash Lite$0.30/$2.50GPT-5.6 Luna$0.20/$1.20Open comparison
Challengers against the big labs
DeepSeek, Qwen, Kimi, GLM and Meta models, priced against each other and against the closed frontier models they compete with.
- DeepSeek V4.1 Flash vs V4 ProThe cheaper model now leads on coding benchmarksDeepSeek V4.1 Flash$0.15/$0.60DeepSeek V4 Pro$0.66/$1.98Open comparison
- Muse Spark 1.3 vs DeepSeek V4.1 FlashClosed mid-price model against a cheap open oneMuse Spark 1.3$1.25/$4.25DeepSeek V4.1 Flash$0.15/$0.60Open comparison
- DeepSeek V4 Pro vs GPT-6 SolReasoning workload comparison, after OpenAI halved its priceDeepSeek V4 Pro$0.66/$1.98GPT-6 Sol$2.00/$10.00Open comparison
- Qwen3.8 Max vs GPT-6 SolSame input price now, different tokenizerQwen3.8 Max$2.00/$6.00GPT-6 Sol$2.00/$10.00Open comparison
- Kimi K3 vs Claude Fable 5.1Open-weight challenger against the frontierKimi K3$3.00/$15.00Claude Fable 5.1$10.00/$50.00Open comparison
- GLM-5.3 vs DeepSeek V4 ProLow-cost open-weight comparisonGLM-5.3$1.40/$4.40DeepSeek V4 Pro$0.66/$1.98Open comparison
Looking for an older model?
GPT-4o, GPT-4o mini, o1 and DeepSeek R1 are retired. These pages compare the models that replaced them.
- Claude vs GPT token costWhat replaced GPT-4o in the balanced tierClaude Sonnet 5$2.00/$10.00GPT-5.6 Terra$2.00/$12.00Open comparison
- DeepSeek vs OpenAI reasoning costSuccessors to R1 and o1, priced on outputDeepSeek V4.1 Flash$0.15/$0.60GPT-6 Astra$10.00/$50.00Open comparison
- Gemini Flash vs GPT miniThe cheap tier after GPT-4o miniGemini 3.8 Flash$0.75/$3.75GPT-5.6 Luna$0.20/$1.20Open comparison
Image models, priced per image
Image APIs bill per image or per image token, so these pages price generated images rather than text. How image generation APIs bill →
- Image generation API pricingCost per image for GPT Image 2.5, Nano Banana, Grok Imagine, Seedream and MAI-ImageOpen comparison
- GPT Image 2.5 vs Nano Banana 2Per-token image billing from OpenAI and Google, priced per imageOpen comparison
- GPT Image 2.5 vs Grok Imagine Image 2.0Quality tiers against a flat $0.04 per imageOpen comparison
What changed this month
Compare approaches, not just models
- Dots vs Grok Bot vs MuseThree agent products compared, and why agent work costs more than chatRead the guide
- The same model at five pricesBatch, flex, fast, ultrafast and peak-hour rates comparedRead the guide
- Prompt caching costWhat a cached prefix saves on a 20-turn agent loopRead the guide
- Long context vs RAG costStuffing a 1M window against retrieval, priced per queryRead the guide