Claude Opus 5.5: four fakes, six facts and every source
Every claim below was checked on 29 October 2026 on Anthropic's own pages: the announcement, the model page, the price list and the Claude Code docs. The links are at the bottom, so you can check today's version yourself.
Part 1 open with your accountFull program 6 modulesTime 5 minutes to read, part 1 about 1 hour
Each claim with a clear verdict: fake, half true or fact
The source for every verdict, on Anthropic's own pages
Part 1: how to pick model and effort in Claude Code and spend less of your plan
A prompt that checks any AI claim against official pages
You need
Nothing for the facts, they are open to everyone
For part 1: Claude Code with a Claude plan that includes it, or an Anthropic API key
Four fakes
Fake"Opus 5.5 is free and unlimited."
The Free plan on claude.com/pricing has Sonnet and Haiku, not Opus. And Anthropic writes that every plan has usage limits, which reset on a rolling five hour window.
Fake"It came out on 26 September."
Anthropic announced Claude Opus 5.5 on 22 September 2026.
Fake"It made every other model useless."
In Anthropic's own benchmark table, GPT-6 Astra scores higher on two tests: AutomationBench (41.4% against 40.0%) and Terminal-Bench-Science (64.6% against 58.7%, the Astra figure as reported by OpenAI). Anthropic also writes that Opus 5.5 performs at the level of Fable 5.1 "on most work", which is not the same as all work.
Half true"It's 40% cheaper."
The price per token dropped 20%: 4 dollars in and 20 dollars out per million tokens, against 5 and 25 for Opus 5. The 40% is Anthropic's estimate of what typical workloads cost to run at default settings, not a price cut.
Six facts
Fact1 million tokens of context.
That is roughly 555 000 words, according to the model page.
FactUp to 128 000 tokens in one answer.
The maximum output is 128K tokens. Through the Batch API Anthropic allows more with a beta header.
Fact4 dollars per million tokens in, 20 out.
That is the API price on the pricing page. On a Claude plan you pay the plan, not per token.
FactIt is the default model in Claude Code.
On Pro, Max, Team, Enterprise and the Anthropic API, Claude Code defaults to Opus 5.5.
FactOn medium effort it matches or beats Opus 5 on high.
That is in Anthropic's own tests on coding and knowledge work, as written in the Claude Code docs.
FactA launch customer kept an agent on one task for more than 18 hours.
In the announcement, Clio describes an engineering task across six repositories that ran unattended overnight and stayed on task for over 18 hours.
Why fakes spread so fast
A new model comes out, and within hours videos and posts repeat each other's numbers. A date moves by four days, a cost estimate turns into a price cut, and a plan with no fee turns into a model with no limits. None of this is checked against the source, because the source is boring: a price table and a docs page.
The fix takes two minutes. Open Anthropic's announcement, the model page and the price list, and read the sentence the claim is based on. Part 1 below includes a prompt that does this check for you in Claude Code.
Sources behind the whole video series
The rules I follow when I publish these videos (AI labels, original uploads, captions), with the date of each page:
No. On Anthropic's pricing page the Free plan has Sonnet and Haiku. Opus needs a paid plan such as Pro or Max, or the Anthropic API where you pay per token. Every plan has usage limits.
Where can I check the price per token myself?
On the pricing page of the Claude Platform docs, platform.claude.com/docs/en/about-claude/pricing. On 29 October 2026 it listed Opus 5.5 at 4 dollars per million input tokens and 20 dollars per million output tokens.
What does 1 million tokens mean in words?
Anthropic's model page puts it at roughly 555 000 words or 2.5 million characters. That is the limit of what Opus 5.5 can read at once in one conversation, not a target you need to fill.
How do I know this page is still right?
Every verdict links to the Anthropic page it comes from, and the check date is on the page. If a page changes, the link shows the new version. I update this page when I notice a change.
I have worked with ads on Google, Meta and TikTok for more than seven years and personally managed ad budgets from $1,000 a day up to $50,000 a month. For the last two years I have built websites, automations and AI assistants with AI every day. I help local businesses in Oslo and on Romerike get more calls and bookings. LinkedIn
