Claude vs OpenAI API Pricing: Which Is Cheaper?
Short answer: Claude vs OpenAI API pricing is now a tie at the top and in the middle: Claude Fable 5.1 and GPT-6 Astra both cost $10/$50 per million input/output tokens, and Claude Sonnet 5 and GPT-6 Sol both cost $2/$10. OpenAI is ten times cheaper at the small end (GPT-6 Luna vs Claude Haiku 4.5). Claude is cheaper for prompts over 272K tokens. Both give 50% off batch jobs.
Prices below come from OpenAI’s API pricing page and Anthropic’s pricing page, checked on September 26, 2026. Both change often.
Claude vs OpenAI API pricing, model by model
| Claude model | In / Out (per 1M) | Closest OpenAI model | In / Out (per 1M) | Difference |
|---|---|---|---|---|
| Claude Fable 5.1 | $10 / $50 | GPT-6 Astra | $10 / $50 | Same list price |
| Claude Opus 5.5 | $4 / $20 | GPT-5.6 Sol | $4 / $20 (promotional) | Same list price |
| Claude Sonnet 5 | $2 / $10 | GPT-6 Sol | $2 / $10 | Same list price |
| Claude Haiku 4.5 | $1 / $5 | GPT-6 Luna | $0.10 / $0.50 | OpenAI 90% cheaper |
Notes on the pairings:
- OpenAI says GPT-5.6 Sol’s $4/$20 promotional price runs at least through November 21, 2026. Opus 5.5 has no exact GPT-6 counterpart; it sits between GPT-6 Sol and GPT-6 Astra.
- Context windows: Claude Fable 5.1, Opus 5.5 and Sonnet 5 have 1M tokens and Haiku 4.5 has 200K. GPT-6 models have a 1.05M window with up to 922K input tokens.
- Anthropic says Claude models from 4.7 on use a tokenizer that produces about 30% more tokens for the same text than older Claude models, and every vendor splits text differently. Equal rate cards don’t guarantee equal bills.
What does each model cost per task?
These are estimates, not measurements: each row assumes a fixed number of input and output tokens and applies list prices with no caching. Real tasks vary a lot, and the two vendors’ tokenizers count the same text differently.
| Task (assumed tokens in / out) | Claude Sonnet 5 | GPT-6 Sol | Claude Haiku 4.5 | GPT-6 Luna |
|---|---|---|---|---|
| Code review, 200 lines (3K / 800) | $0.014 | $0.014 | $0.007 | $0.0007 |
| Bug fix, one function (5K / 1K) | $0.020 | $0.020 | $0.010 | $0.001 |
| Feature, ~200 lines (20K / 6K) | $0.10 | $0.10 | $0.05 | $0.005 |
| Chat, 10 turns (30K / 5K) | $0.11 | $0.11 | $0.055 | $0.006 |
| Long document, 400K-token prompt (400K / 3K) | $0.83 | $1.65 | Too long (200K limit) | $0.08 |
At normal prompt sizes the mid-tier models cost exactly the same, so the real difference is how many tokens and retries each needs on your work. The long-document row is where pricing rules diverge: past 272K input tokens OpenAI doubles the input rate and raises output by half for the whole request, while Claude stays flat.
How do batch, caching and long-prompt pricing differ?
| Feature | Claude | OpenAI |
|---|---|---|
| Batch API (async) | 50% off input and output | 50% off input and output |
| Flex processing (slower) | — | 50% off on supported models |
| Cache reads | $0.20 on Sonnet 5 and Opus 5.5, $0.25 on Fable 5.1, $0.10 on Haiku 4.5 | 10% of input: $0.20 on GPT-6 Sol, $1.00 on GPT-6 Astra, $0.01 on Luna |
| Cache writes | 1.25× input for a 5-minute cache, 2× for 1 hour | 1.25× input on GPT-6 models |
| Prompts over 272K tokens | No surcharge on Claude 4.6 and later (up to 1M) | 2× input and cache rates, 1.5× output |
Older versions of this post said Claude had no batch option. That was wrong: both vendors take 50% off batch jobs, and Anthropic’s discounts for batch and caching can be combined.
The caching row matters most at the top tier. If your app re-sends a large system prompt or codebase, Fable 5.1 reads cached tokens at $0.25 per million against $1.00 on GPT-6 Astra, which can outweigh everything else on an input-heavy workload.
Which provider is better for specific use cases?
- Daily coding assistant: a tie on price between Claude Sonnet 5 and GPT-6 Sol; pick on output quality for your stack.
- High-volume simple tasks (classification, formatting, extraction): GPT-6 Luna, at a tenth of Haiku 4.5’s price.
- Long documents and big codebases in one prompt: Claude, with flat pricing up to 1M tokens.
- Input-heavy apps at the top tier: Claude Fable 5.1, for its cheaper cache reads.
- Batch processing: either; both are 50% off.
- Non-urgent work you can’t batch: OpenAI Flex, 50% off with slower responses.
Many teams use both: GPT-6 Luna for cheap bulk work and Claude or GPT-6 Sol for the tasks that need a stronger model. If you’re weighing a third provider, see is Gemini cheaper than Claude? and the full LLM API pricing comparison.
How do you track costs across both providers?
For API spend, each vendor’s console is the source of truth: the Claude Console and the OpenAI dashboard both break usage down by model and day. Check them weekly while you tune which model handles which task.
If you use the subscription tools instead (Claude Code on a Pro or Max plan, Codex through ChatGPT), FavTray’s free AI Usage Tracker shows both plans’ usage limits side by side in the Mac menu bar. For a wider look at tracking tools, including cloud options for teams, see our AI usage tracking tools comparison; for Claude specifically, our Claude API cost tracking guide. Deciding between a plan and the API? See whether Claude Max is worth it.
For ways to cut costs regardless of provider, see our guide to reducing AI API costs, and for subscription coding tools, our AI coding assistant pricing comparison.
Frequently Asked Questions
Is Claude cheaper than OpenAI?
At the top and middle tiers the list prices are identical: Claude Fable 5.1 and GPT-6 Astra both cost $10 input and $50 output per million tokens, and Claude Sonnet 5 and GPT-6 Sol both cost $2 and $10. OpenAI is far cheaper at the small end (GPT-6 Luna at $0.10/$0.50 against Claude Haiku 4.5 at $1/$5). Claude is cheaper for prompts over 272K tokens. Prices checked September 26, 2026.
Which is cheaper for coding tasks, Claude or GPT?
Claude Sonnet 5 and GPT-6 Sol have the same per-token price, including cached input, so the cheaper one is whichever finishes your task in fewer tokens and fewer retries. That depends on your codebase and prompts, so run the same 20 real tasks through both and compare the bills.
Does Claude have a Batch API like OpenAI?
Yes. Anthropic's Batch API takes 50% off input and output tokens for asynchronous jobs, the same discount as OpenAI's Batch API. OpenAI also offers Flex processing at 50% off for slower, non-urgent requests on supported models.
Which is cheaper for long documents?
Claude. Sonnet 5, Opus 5.5 and Fable 5.1 bill their full 1M-token window at the standard rate. OpenAI's GPT-6 models charge 2× input and 1.5× output for the whole request once a prompt passes 272K tokens, so a 400K-token prompt costs about twice as much on GPT-6 Sol as on Sonnet 5.
What is the cheapest model from either company?
OpenAI's. GPT-6 Luna costs $0.10 input and $0.50 output per million tokens, and older models such as gpt-5-nano ($0.05/$0.40) are cheaper still. Claude's cheapest current model is Haiku 4.5 at $1/$5, with a 200K-token context window.