Claude Opus 5.5 fast mode costs a third less than GPT-6.1 Sol Ultrafast, but you need an invite

Claude Opus 5.5 in fast mode costs a third less than GPT-6.1 Sol Ultrafast on input, output and cached input, going by both companies' own pricing pages, read this morning. The catch is access: Sol Ultrafast is open to every OpenAI API user, while Anthropic's fast mode is a research preview you request through an account manager or a waitlist.

THE PRICE, PER MILLION TOKENS

Claude Opus 5.5 fast mode (Anthropic's pricing page): $8 input, $40 output, $0.40 cached input, $10 for 5-minute cache writes and $16 for 1-hour writes. That is exactly double Opus 5.5 at standard speed ($4 / $20 / $0.20 / $5), and Anthropic charges the same rate across the whole context window.

GPT-6.1 Sol Ultrafast (OpenAI's pricing page): $12 input, $60 output, $0.60 cached input, $15 cache writes, for prompts up to 272K tokens. Above that it is $24 / $90 / $1.20 / $30. Sol at standard speed is $2 / $10, so Ultrafast is six times the price.

Side by side, Opus fast mode is 33% cheaper on input, output, cached input and 5-minute cache writes, and two thirds cheaper on input once a Sol prompt passes 272K tokens. Anthropic's 1-hour cache write ($16) is the one line that costs more than OpenAI's ($15).

WHO CAN USE IT

Anthropic: research preview on the Claude API only, not on Amazon Bedrock, Google Cloud or Microsoft Foundry, not with the Batch API or a Priority Tier commitment, and with its own rate limits. Switching between fast and standard speed throws away the prompt cache. Fast mode also runs on Opus 5 and Opus 4.8, at $10 / $50.

OpenAI: its docs say Ultrafast for GPT-6.1 Sol and GPT-6 Astra is "available to all API users", set with service_tier "ultrafast". In Codex and ChatGPT Work it is limited to Pro 500 and some Enterprise and Edu plans.

HOW FAST

Both speed claims are the vendors' own and use different baselines. Anthropic says up to 2.5x the output tokens per second of Opus 5.5 at standard speed; OpenAI called Ultrafast up to 8x the speed of Sol Standard when it rolled out on October 8. @kimmonismus put both at roughly 280 to 300 tokens per second on X; we have not seen an independent measurement of either.

WHAT WE DO NOT KNOW

Whether the two are equally fast in practice, and how quickly Anthropic is letting people in from the waitlist.