Comparison
Both providers sell access to large language models per token, priced separately for input and output, with discounts for cached input. This comparison puts the published rates for their current flagship, mid-tier and smallest listed models side by side so you can estimate what a workload would cost on each.
Last verified
Developers and product teams building on an LLM API who need to estimate and compare running costs, and anyone deciding which provider to prototype with first.
On small screens, scroll the table sideways to see every product.
| Criterion | ||
|---|---|---|
| Pricing (verified)From each tool's verified plans, with the date SIDAN checked them. | Usage-based pricingChecked October 24, 2026 | Usage-based pricingChecked October 24, 2026 |
| Pricing model | Usage-based, per 1M tokens, input and output priced separately | Usage-based, per 1M tokens, input and output priced separately |
| Flagship model (as listed 24 Oct 2026) | gpt-6-astra: USD 10 input / USD 50 output | Claude Fable 5.1: USD 10 input / USD 50 output |
| Mid-tier model | gpt-6.1-sol: USD 2 input / USD 10 output | Claude Opus 5.5: USD 4 input / USD 20 output |
| Smallest listed model | gpt-6-luna: USD 0.10 input / USD 0.50 output | Claude Sonnet 5.5: USD 2 input / USD 10 output |
| Cached input | USD 1.00 / 0.10 / 0.01 per 1M cached input tokens (astra / sol / luna) | Cache hits USD 0.25 / 0.20 / 0.10 per 1M tokens (Fable / Opus / Sonnet); cache writes priced separately |
| Billing notes | Standard tier rates for requests up to 272K context; longer contexts cost more | All prices in USD per the page; some models limited availability or retired |
| Best for | Lowest listed small-model rate; OpenAI tooling | Agents and long prompts that benefit from caching |
OpenAI's developer API for calling its language models, billed per million tokens with separate input, cached-input and output rates.
Three clearly tiered models on the page; cached-input rates published; very low-cost small model
Long-context requests (above 272K tokens) cost more than the standard rates; output tokens cost several times input
Teams whose stack and tooling already centre on OpenAI, or who need the lowest listed small-model rate
Anthropic's developer API for the Claude model family, billed per million tokens with separate input, output and prompt-caching rates.
Cache-read rates published per model; mid-tier model priced below the OpenAI mid-tier at the time of checking
Some listed models are limited availability or retired; output tokens cost several times input
Teams building agents or long-prompt applications who can benefit from prompt caching
You are prototyping and cost is secondary: pick the provider whose SDK and tooling fit your stack; both offer a small model priced well under USD 1 per million input tokens for experiments.
Your workload is output-heavy (long generations): output tokens cost several times more than input on both providers, so compare output rates first and measure your actual input/output ratio before choosing.
You re-send long prompts or system instructions: cached-input pricing matters more than headline rates; both providers price cache reads far below base input, with different cache-write rules on Anthropic's side, so read each page's caching section.
You need a specific capability (context length, tool use, vision, fine-tuning): check the model documentation rather than price alone; a cheaper token is not cheaper if you need more of them.
At the time of checking, the two flagship models were listed at the same base rates (USD 10 input / USD 50 output per 1M tokens), while mid-tier and small models differ. The decision usually comes down to model fit, tooling and caching behaviour for your prompts rather than list price alone. Rates change often; both pricing pages are linked below with the date we checked them.
Research-based comparison. Every figure comes from the vendors' official pricing and documentation pages, recorded as plans in the SIDAN Stack catalog with the date checked; the pricing row above is generated from those verified records and shows currency, unit and billing period exactly as published. We have not tested these products hands-on. Vendor claims are attributed; our own assessment is marked as such. Where a page did not state something plainly, the cell says so instead of guessing. Token rates are per 1 million tokens in USD as stated on each provider's pricing page. Rates for long-context requests, batch processing and other tiers are not included; check each page for those.
This comparison is research-based: it is built from the vendors' official pages, documentation and the sources listed below. SIDAN Stack has not tested these products hands-on unless the text says so, and nothing here is scored or ranked. How we research.
To compare AI API pricing, estimate input, output and cached-input tokens per request, copy each model's per-million-token rates with their tier and context band, multiply by monthly volume, apply only documented batch and caching discounts, and check the result against the provider's rate limits and spend caps. This tutorial walks through the method with verified OpenAI and Anthropic rates as listed on 24 October 2026 and an illustrative worked example.
Last verified . Vendors change their products and prices; confirm details on the official pages before buying.
