OpenAI vs Claude API pricing: a worked cost example.
In OpenAI vs Claude API pricing, list prices were identical at several price points on 5 November 2026, including $0.10 in and $0.50 out, and $2 in and $10 out per million tokens. So your bill will depend more on the model tier, caching and token use than on the provider. Test both with your own examples. Quality wasn’t tested here.
Reviewed by Anubhav Gupta, Bespoke Devs · Updated
OpenAI vs Claude API pricing: the worked example
Using made-up numbers: 10,000 requests a month, with 2,000 input and 500 output tokens each. That’s 20 million tokens in and 5 million out. Models are paired by price only, which says nothing about quality.
| Price point (in / out per 1M) | Claude model | OpenAI model | Example month, each |
|---|---|---|---|
| $0.10 / $0.50 | Haiku 5.5 (prompts up to 100,000 tokens) | gpt-6-luna | $4.50 |
| $1 / $5 | Haiku 4.5 | gpt-6.1-terra | $45 |
| $2 / $10 | Sonnet 5.5 | gpt-6.1-sol | $90 |
| $4 / $20 | Opus 5.5 | gpt-5.6-sol | $180 |
| $10 / $50 | Fable 5.1 | gpt-6-astra | $450 |
Standard list prices. OpenAI’s figures are its short-context rates, for up to 272K input tokens. Both providers list many other models.
Where OpenAI vs Claude API pricing differs: repeated text
Both providers charge less for prompt text they’ve already processed. Here’s what each charges per million cached tokens read.
| Price point | Claude cache read | OpenAI cached input | Lower |
|---|---|---|---|
| $0.10 / $0.50 | $0.01 | $0.01 | Same |
| $1 / $5 | $0.10 | $0.05 | OpenAI |
| $2 / $10 | $0.10 | $0.10 | Same |
| $4 / $20 | $0.20 | $0.40 | Claude |
| $10 / $50 | $0.25 | $1.00 | Claude |
Anthropic also lists a cache write price, which is above its normal input price: $2.50 per million for a five-minute write on Sonnet 5.5, against $2 for normal input. OpenAI’s pricing table lists no separate write price. So heavy cache use could tip either way, depending on how often your cache gets reused.
The same example with caching
Say 1,500 of your 2,000 input tokens repeat in every request, such as a long system prompt. On either $2 / $10 model, that’s 15 million cached tokens at $0.10 ($1.50), 5 million normal tokens at $2 ($10) and 5 million output tokens at $10 ($50). The total is $61.50, down from $90. On Claude, add the cost of writing to the cache.
Other pricing details to check
List prices aren’t the whole OpenAI vs Claude API cost. Check these as well.
- Batch discountAnthropic’s Batch API gives 50% off. OpenAI’s batch prices are about half of standard on most models. For the $2 / $10 example, that’s $45 on either one.
- Long promptsAnthropic says Claude 4.6 and later models, except Haiku 5.5, include the full 1M token context at standard pricing. Haiku 5.5 costs more above 100,000 tokens. OpenAI’s rates above are for up to 272K input tokens, so check its long-context rates.
- Data locationAnthropic charges 1.1 times the normal rate for US-only inference on Claude 4.6 and later. If you need this on either provider, check the cost in its current docs.
- ToolsServer-side tools such as web search can add charges beyond tokens. Compare them if your feature uses them.
- Older modelsOlder models on both price lists can cost more than newer ones. Don’t assume that older means cheaper.
How to choose
Price could decide it if
- Your volume is high and the output is simple
- Your feature relies heavily on caching, so compare the rates above
- You’ve tested both, and quality is close
Other things should decide it if
- One model gets your examples right more often
- A wrong answer costs more than the tokens do
- You need a specific feature or data terms
- You already use one provider’s cloud or tools
How to test fairly
Take 20 to 50 real examples and give both models the same prompt. Score the results before you know which is which. Note the tokens used, because models can use different amounts for the same task, and that changes the real bill. Then pick the cheaper one that passes. Keep the model behind a thin layer in your code, so that you can switch when prices or models change. They change often.
We build with Claude, OpenAI or Gemini, depending on the job. See our notes on AI MVP development if you’d like us to do this for you.
Common questions
At the price points above, neither on standard input and output. With caching, OpenAI was lower at $1 / $5, and Claude was lower at $4 / $20 and $10 / $50. Real cost depends on the model you need and how many tokens it uses for your task.
Gemini prices sit in a similar range, and some are changing soon. On 5 November 2026, gemini-3.1-flash-lite was $0.25 in and $1.50 out, and gemini-4-argon was $2 and $10 through 30 November, then $4 and $20. See our [cost comparison](/ai-mvp-development/questions/what-do-llm-features-cost).
Anthropic’s privacy centre says that by default it doesn’t use inputs or outputs from commercial products for training, unless you opt in, for example by sending feedback. Check OpenAI’s current enterprise privacy page directly for its terms.
We haven’t tested that here, and it depends on the task. Run your own examples through both. General benchmarks won’t tell you what your feature needs.
Keep reading
Tell us what you’re building.
Takes under a minute. We come back with a scope and a fixed price.
- Reply within 24 hoursA real person reads every brief.
- Fixed scope, fixed priceYou get a written proposal, no hourly billing.
- You own the codeFull ownership of code and assets at launch.