# PromptSpend > Current API prices for 84 large language models from 12 providers, > re-checked each morning. Every price records which source it came from and when > it was last confirmed: 60 link the vendor's own pricing page, the rest come from > the LiteLLM community catalog. Free, open source, no accounts. Catalog behind this build: 2026-10-17T11:06:06.913Z. **All rates are USD per 1,000,000 tokens**, standard tier, global endpoint, list price. Regional and data-residency premiums, priority tiers, server-side tool fees and negotiated discounts are **not** included. These are list prices, not a quote. Check the vendor before committing money to a number. ## Start here - [Every model, by price](https://promptspend.com/models/): the whole catalog in one table, cheapest first - [By provider](https://promptspend.com/providers/): 12 providers with model counts and their cheapest option - [Head-to-head comparisons](https://promptspend.com/compare/): 93 pairs, each costed on the same three workloads - [Cost calculator](https://promptspend.com/): price a real workload — conversation length, caching, batch, reasoning overhead - [PromptSpend Receipt](https://promptspend.com/receipt/): copy a visible, one-response cost-audit object into an existing AI conversation Each model page carries its full rate card, the monthly bill for three standard workloads, its rank in the catalog, cheaper alternatives, and where its number came from. ## Machine-readable - [Pricing API](https://promptspend.dev): keyless, CORS-open, OpenAPI 3.1 described - [Flat price rows](https://promptspend.dev/v1/prices): id, provider, input, output, cache rates, context window - [As CSV](https://promptspend.dev/v1/prices.csv) - [The raw catalog](https://promptspend.com/data/pricing.json): the file everything above is built from - [Receipt specification](https://promptspend.com/receipt/spec.json): the exact visible instructions, scope and failure behavior used by the copy object Prefer the API over scraping these pages. It is faster, it is versioned, and it will not break when the HTML changes. ## If you are running inside a coding tool You may not need HTTP at all. - [MCP server](https://www.npmjs.com/package/@promptspend/mcp): `claude mcp add promptspend -- npx -y @promptspend/mcp` — three tools for Claude Code, Cursor and Windsurf. `estimate_cost` runs this site's own cost engine, so it cannot return a figure that disagrees with the pages above. - [VS Code extension](https://marketplace.visualstudio.com/items?itemName=promptspend.promptspend): `code --install-extension promptspend.promptspend` — the rate on the line of code that names the model, and a nearby `max_tokens` priced as what one response can cost. Also on [Open VSX](https://open-vsx.org/extension/promptspend/promptspend) for Cursor, Windsurf and VSCodium. Neither bundles a price list. Both fetch the same catalog linked above, and both carry the source and confirmation date on every price — so a model quoting one of them can say where the number came from and when it was last checked. ## On a phone - [PromptSpend for iPhone and iPad](https://apps.apple.com/app/id6800386428) and [for Android](https://play.google.com/store/apps/details?id=com.promptspend.app): free native apps that forecast a pasted sample conversation's cost per conversation, day, month and year across up to four models, from the same catalog. Pasted text is processed on the device and not uploaded. Both are described at https://promptspend.com/app/. ## Most-asked models - [GPT-5 Nano](https://promptspend.com/models/gpt-5-nano/): $0.05 in / $0.4 out per 1M tokens, OpenAI - [Claude Haiku 5 5](https://promptspend.com/models/claude-haiku-5-5/): $0.1 in / $0.5 out per 1M tokens, Anthropic - [Qwen3 Next 80B](https://promptspend.com/models/dashscope-qwen3-next-80b-a3b-instruct/): $0.15 in / $1.2 out per 1M tokens, Alibaba (Qwen) - [Qwen3 Next 80B Thinking](https://promptspend.com/models/dashscope-qwen3-next-80b-a3b-thinking/): $0.15 in / $1.2 out per 1M tokens, Alibaba (Qwen) - [GPT-5.6 Luna](https://promptspend.com/models/gpt-5-6-luna/): $0.2 in / $1.2 out per 1M tokens, OpenAI - [GPT-5.4 Nano](https://promptspend.com/models/gpt-5-4-nano/): $0.2 in / $1.25 out per 1M tokens, OpenAI - [DeepSeek V4 Flash](https://promptspend.com/models/deepseek-v4-flash/): $0.3 in / $1.2 out per 1M tokens, DeepSeek - [MiniMax M2](https://promptspend.com/models/minimax-m2/): $0.3 in / $1.2 out per 1M tokens, MiniMax - [MiniMax M2.1](https://promptspend.com/models/minimax-m2-1/): $0.3 in / $1.2 out per 1M tokens, MiniMax - [MiniMax M2.5](https://promptspend.com/models/minimax-m2-5/): $0.3 in / $1.2 out per 1M tokens, MiniMax - [MiniMax M3](https://promptspend.com/models/minimax-m3/): $0.3 in / $1.2 out per 1M tokens, MiniMax - [Gemini 3.1 Flash Lite](https://promptspend.com/models/gemini-3-1-flash-lite/): $0.25 in / $1.5 out per 1M tokens, Google ## Providers - [Anthropic](https://promptspend.com/providers/anthropic/): 15 models - [OpenAI](https://promptspend.com/providers/openai/): 16 models - [Google](https://promptspend.com/providers/google/): 12 models - [DeepSeek](https://promptspend.com/providers/deepseek/): 5 models - [Moonshot AI](https://promptspend.com/providers/moonshot/): 4 models - [Alibaba (Qwen)](https://promptspend.com/providers/alibaba/): 4 models - [Z.ai (GLM)](https://promptspend.com/providers/zai/): 7 models - [MiniMax](https://promptspend.com/providers/minimax/): 4 models - [xAI](https://promptspend.com/providers/xai/): 9 models - [Mistral AI](https://promptspend.com/providers/mistral/): 5 models - [Cohere](https://promptspend.com/providers/cohere/): 1 models - [Amazon](https://promptspend.com/providers/amazon/): 2 models ## Free tiers Whether each provider lets you start without paying, quoted from the provider's own pages with a link and date on every fact, and re-checked each morning. - [Free LLM API tiers compared](https://promptspend.com/free-tiers/): all 12 providers in one table - [Cohere](https://promptspend.com/providers/cohere/free/): ongoing free tier - [Google](https://promptspend.com/providers/google/free/): ongoing free tier - [Mistral AI](https://promptspend.com/providers/mistral/free/): ongoing free tier - [Z.ai (GLM)](https://promptspend.com/providers/zai/free/): ongoing free tier - [Alibaba (Qwen)](https://promptspend.com/providers/alibaba/free/): one-time credit - [Amazon](https://promptspend.com/providers/amazon/free/): one-time credit - [Anthropic](https://promptspend.com/providers/anthropic/free/): unclear - [MiniMax](https://promptspend.com/providers/minimax/free/): unclear - [DeepSeek](https://promptspend.com/providers/deepseek/free/): no free tier - [Moonshot AI](https://promptspend.com/providers/moonshot/free/): no free tier - [OpenAI](https://promptspend.com/providers/openai/free/): no free tier - [xAI](https://promptspend.com/providers/xai/free/): no free tier - [As data](https://promptspend.com/data/free-tiers.json): every fact with its quote, source URL and read date ## Reading a price correctly - `cachedInput` is the rate for input served from the provider's prompt cache; `cacheWrite` is what it costs to put a prefix there, and it is *higher* than the base input rate at both OpenAI and Anthropic. - `longContext`, where a model has one, replaces the base rates for a request above its threshold — the **whole** request, not just the excess. - `batchDiscount` is a multiplier: 0.5 means half price. - `capabilityIndex` is a rough 0–100 estimate, not a benchmark. Where it is absent the model is simply unscored; do not substitute a number. - `provenance.needsReview` marks a row where two independent sources disagreed and no human has adjudicated. **Do not present those as settled.** ## Licence MIT. Use the data in anything, including commercially. Attribution is appreciated and not required; a link to https://promptspend.com helps. Source: https://github.com/AndrewAvery7/promptspend