OpenAI vs Anthropic API pricing 2026, model by model

If you are choosing between OpenAI and Anthropic, or already pay one and wonder whether the other is cheaper, the headline rates will not settle it. In three of the four price tiers, the two providers now charge exactly the same per million tokens. What separates them is the fine print: long-prompt surcharges, cache prices and how long a cache lasts. This page lists both price lists tier by tier and shows, with worked numbers, where the bills come apart. All prices are list prices in US dollars per million tokens on the standard pay-as-you-go tier, checked on each provider's own pricing page on 27 October 2026.

Price table, tier by tier

Anthropic also lists Claude Mythos 5.1 at the same prices as Fable 5.1. OpenAI's batch prices are 50% of standard, as stated on each model page; Anthropic lists its batch prices directly. Both providers bill reasoning or thinking tokens as output, so a model that thinks at length on a simple task costs more than its rates suggest.

The three matched pairs

GPT-6 Astra vs Claude Fable 5.1

Same input, output and batch rates, and the same 30-minute and 5-minute cache write price of $12.50. The difference is the cache read: $1.00 per million on Astra, $0.25 on Fable 5.1. On a workload that rereads a large fixed prompt, Fable 5.1's input bill is a quarter of Astra's. Astra also pays the long-prompt surcharge above 272K tokens; Fable 5.1 does not.

GPT-6.1 Sol vs Claude Sonnet 5.5

Identical on every line of the table: $2 input, $0.10 cached, $2.50 for the short cache write, $10 output, half price on batch. On ordinary requests the bills will differ only by the number of tokens each model uses. Prompts over 272K tokens are where they split, covered below.

GPT-6 Luna vs Claude Haiku 5.5

Both list at $0.10 input, $0.01 cached and $0.50 output, the lowest rates either provider offers. They differ on long prompts: Haiku 5.5's rates rise five-fold above 100,000 prompt tokens, while Luna's input rate doubles and its output rate rises 1.5 times only above 272,000.

Models without a match

Claude Opus 5.5, at $4 input and $20 output, sits between the two OpenAI models that bracket it: 40% of GPT-6 Astra's rates and twice GPT-6.1 Sol's. Our flagship cost-per-task comparison works through when Astra or Opus 5.5 comes out cheaper.

GPT-6.1 Terra, released on 26 October 2026, lists at $1 input, $0.05 cached and $5 output, half of GPT-6.1 Sol. OpenAI describes it as a fast, efficient model for visual reasoning, computer use and writing. Anthropic's current line-up has nothing at that price point; the nearest is the previous-generation Claude Haiku 4.5, which lists at the same $1 and $5 but $0.10 for cache reads.

Long prompts: the biggest difference

OpenAI's model pages for GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna all say that prompts with more than 272K input tokens are priced at 2x the input and cache rates and 1.5x the output rate for the full request. Anthropic's pricing page says Claude 4.6 and later models, except Haiku 5.5, include the full 1M-token context window at standard prices, and that Haiku 5.5 requests whose prompt is over 100,000 tokens, counting cache reads and writes, pay higher prices: $0.50 input and $2.50 output.

No caching, standard tier, equal token counts on both sides; these are illustrations, not measurements. The pattern holds whatever the volume: for prompts between 100K and 272K tokens the small OpenAI model is far cheaper, and for prompts over 272K tokens the Anthropic mid and flagship models are. If you send whole documents, codebases or long chat histories, find out how many of your requests cross each line before comparing anything else. Retrieving only the relevant parts of a document often keeps requests under the threshold on either provider.

Caching: read prices and cache lifetimes

Cache reads cost 10% of the input rate on GPT-6 Astra, GPT-6 Luna and Claude Haiku 5.5, 5% on GPT-6.1 Sol, GPT-6.1 Terra, Claude Opus 5.5 and Sonnet 5.5, and 2.5% on Claude Fable 5.1. Writing to the cache costs 1.25 times the input rate on both providers for the short option. The lifetimes differ: OpenAI's short cache write is listed as 30 minutes and Anthropic's as 5 minutes. For a longer lifetime, Anthropic offers 1 hour at 2 times the input rate and OpenAI, since 20 October 2026, 12 hours at 2.5 times.

Example: a 100,000-token fixed prompt read from cache, plus a 1,000-token answer, costs $150 per 1,000 requests on GPT-6 Astra and $75 on Claude Fable 5.1. On Sol and Sonnet 5.5 the same requests cost an identical $20.

If your traffic has gaps of more than a few minutes, the short Anthropic cache may expire between requests and you pay the write again. Compare what you pay in cache writes against the longer cache option, using your own usage export.

Batch, faster processing, regions and tools

• Batch: half price on both, for work that can wait. OpenAI also lists Flex processing at the batch rate.
• Faster processing: Anthropic lists fast mode for Opus 5.5 at $8 input and $40 output, twice the standard rate, not available with batch. OpenAI lists Fast mode (renamed from Priority on 30 July 2026) and an Ultrafast mode for GPT-6 Astra and GPT-6.1 Sol, both above standard rates; check its pricing page for the current figures.
• Regional processing: OpenAI charges a 10% uplift on data residency endpoints for models released on or after 5 March 2026. Anthropic applies a 1.1x multiplier when you require US-only processing on Claude 4.6 and later models.
• Web search: $10 per 1,000 searches on both, plus the tokens the results add to the prompt.

Older models cost more

Many bills still run on older models that now list higher than their replacements. Anthropic lists Claude Sonnet 4.5 and 4.6 at $3 input and $15 output, against $2 and $10 for Sonnet 5.5, and Claude Opus 4.5 to 4.8 at $5 and $25, against $4 and $20 for Opus 5.5. OpenAI's changelog shows GPT-5.6 Sol at a promotional $4 input and $20 output until at least 21 November 2026, against $2 and $10 for GPT-6.1 Sol. Moving to the current model in the same family is often the simplest saving, once you have checked answer quality on your own examples.

How to compare on your own traffic

1. Find your tier. Most teams compare like with like: Sol with Sonnet 5.5, Luna with Haiku 5.5. Pricing alone rarely justifies switching between matched models.
2. Count long prompts. The share of requests over 100K and over 272K tokens can decide the comparison on its own.
3. Check cache reads and gaps. A high cache share favours the lower read price; bursty traffic favours the longer lifetime.
4. Measure tokens per task. Each provider counts tokens differently and thinks for a different length, so replay 20 to 50 real requests on both and compare the billed tokens, not the rates.

The free API pricing calculator applies every rule on this page, including long-prompt rates and batch, for your request volume and token counts. To see what you spend by model today, run your usage export through the free Bill Analyzer. If you want your real requests priced on every current OpenAI, Anthropic and Google model with a ranked savings plan, that is what the LLM Cost Report delivers.