Claude vs GPT: API Pricing Compared
Side-by-side Claude and GPT API pricing on the same workload, including prompt caching, batch rates and the budget tier Anthropic does not have.
Short answer
OpenAI covers a far wider price range at both ends. Anthropic has no budget tier, but is cheaper than OpenAI at the frontier.
Prices verified 2026-08-15. Set your own token counts below, because the ranking changes with workload shape.
| Model | Input | Output | Monthly total | vs cheapest |
|---|---|---|---|---|
GPT-5 nano OpenAI gpt-5-nano | $13.80 | $6.40 | $20.20 | 1.0x |
GPT-5.6 Luna OpenAI gpt-5.6-luna | $55.20 | $19.20 | $74.40 | 3.7x |
GPT-5.4 nano OpenAI gpt-5.4-nano | $55.20 | $20.00 | $75.20 | 3.7x |
GPT-5 mini OpenAI gpt-5-mini | $69.00 | $32.00 | $101.00 | 5.0x |
GPT-5.4 mini OpenAI gpt-5.4-mini | $207.00 | $72.00 | $279.00 | 13.8x |
Claude Haiku 4.5 Anthropic claude-haiku-4-5 | $276.00 | $80.00 | $356.00 | 17.6x |
GPT-5.1 OpenAI gpt-5.1 | $345.00 | $160.00 | $505.00 | 25.0x |
Claude Sonnet 5 Anthropic claude-sonnet-5 intro rate | $552.00 | $160.00 | $712.00 | 35.2x |
GPT-5.6 Terra OpenAI gpt-5.6-terra | $552.00 | $192.00 | $744.00 | 36.8x |
GPT-4.1 OpenAI gpt-4.1 | $660.00 | $128.00 | $788.00 | 39.0x |
GPT-5.4 OpenAI gpt-5.4 | $690.00 | $240.00 | $930.00 | 46.0x |
Claude Sonnet 4.6 Anthropic claude-sonnet-4-6 | $828.00 | $240.00 | $1,068 | 52.9x |
GPT-4o OpenAI gpt-4o | $1,050 | $160.00 | $1,210 | 59.9x |
Claude Opus 5 Anthropic claude-opus-5 | $1,380 | $400.00 | $1,780 | 88.1x |
Claude Opus 4.8 Anthropic claude-opus-4-8 | $1,380 | $400.00 | $1,780 | 88.1x |
GPT-5.6 Sol OpenAI gpt-5.6-sol | $1,380 | $480.00 | $1,860 | 92.1x |
GPT-5.5 OpenAI gpt-5.5 | $1,380 | $480.00 | $1,860 | 92.1x |
Claude Fable 5 Anthropic claude-fable-5 | $2,760 | $800.00 | $3,560 | 176.2x |
GPT-5.5 Pro OpenAI gpt-5.5-pro | $18,000 | $2,880 | $20,880 | 1033.7x |
Across the models shown, this workload ranges from $20.20 to $20,880 a month, a 1034x spread. Model choice is almost always a larger lever than prompt optimisation.
Prices are USD per million tokens, taken from each provider's own pricing page and last verified on 2026-08-15. Standard tier only: enterprise discounts, committed-use pricing, and free tiers are not modelled. Verify against Anthropic, OpenAI, Google, xAI before committing budget.
Comparing something else? See the other head-to-heads or use the full calculator across all four providers.
How to use Claude vs GPT: API Pricing Compared
Describe one request
Enter the input and output tokens for a single typical request, or pick a preset. The table is fixed to Anthropic and OpenAI.
Set volume and caching
Add your monthly request count, then set the share of input tokens served from the prompt cache. Caching usually moves the answer more than the choice of provider does.
Compare the monthly bill
Models are ranked cheapest first across both providers, with the multiple over the cheapest option so you can see whether a switch is worth the migration.
The gap is at the bottom, not the top
On the same RAG workload, the cheapest Claude model costs about 17 times the cheapest GPT model. That is the single most important number on this page, and it is not a small-print detail: if your requirement is high volume at low cost, Anthropic does not currently field a model that competes. GPT-5 nano and GPT-5 mini exist in a price band Anthropic has left empty.
Flip to the top of each catalogue and the comparison reverses. OpenAI's most expensive model runs roughly six times Anthropic's most expensive one, because OpenAI prices its reasoning-heavy Pro tier far above anything in the Claude range. If you are already committed to frontier-class output, Claude is the cheaper frontier.
So the honest framing is not "which is cheaper". It is that OpenAI's catalogue spans about a thousandfold from floor to ceiling while Anthropic's spans about tenfold, and which one wins depends entirely on where in that range your workload sits.
Where the sticker price stops predicting the bill
Both providers charge roughly a tenth of the input rate for tokens served from a cached prefix, and both make that the largest single lever on a real bill. A coding agent resending the same 100K-token context on every request pays almost nothing for it on the second call onward. Set the cache slider to what your workload actually achieves before drawing conclusions from any row in the table.
Reasoning tokens are the other distortion, and they hit the comparison asymmetrically because the two providers surface thinking differently. On both, thinking tokens bill as output, and output is roughly five times the input rate. A request that returns 200 visible tokens after reasoning at length can bill like a request that returned several thousand.
Neither provider's list price includes committed-use or enterprise discounts. At meaningful volume both negotiate, so treat every figure here as an upper bound rather than a quote.
Frequently Asked Questions
Is Claude cheaper than GPT?
It depends where in each range you look. Anthropic's cheapest model costs about 17 times OpenAI's cheapest, so for high-volume low-cost work GPT wins outright. At the top of both catalogues the order reverses, and Claude is meaningfully cheaper than OpenAI's most expensive tier. Set your real token counts in the calculator above rather than comparing headline rates.
Why is the cheapest Claude model so much more expensive?
Anthropic does not currently ship a nano or micro tier. Its smallest model, Claude Haiku, is positioned against OpenAI's mid-range rather than against GPT-5 nano. That is a product decision rather than a pricing quirk, and it means classification, routing and extraction workloads at scale are usually cheaper on OpenAI.
Does prompt caching change the answer?
Often, yes. Both providers charge about a tenth of the input rate for cached tokens, so a workload with a large stable prefix and a high hit rate can see its input costs fall by 80 to 90 percent. Because the two catalogues sit at different price points, a high cache hit rate compresses the gap between them considerably.
Which has the larger context window?
They are close enough that it rarely decides anything. The largest published Claude window is 1,000,000 tokens and the largest published GPT window is 1,050,000. Retrieval quality degrades well before either limit, so the practical constraint is almost never the published number.
Suggested Tools
LLM API Cost Calculator
Compare what a workload costs across Anthropic, OpenAI, Google and xAI, including prompt caching, batch rates and long-context tiers.
Token Counter & Context Visualizer
Paste a prompt or file to estimate its token count, see how much of each model's context window it fills, and what it costs to send.