Next Claude Opus Released: Opus 5.5 Lands At $4 Per Million

The Quick Answer

Anthropic released Claude Opus 5.5 on Tuesday at $4 per million input tokens, sixty percent under the flagship it shipped three weeks ago. The company says the new model "performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5." Those are two different baselines in one sentence, and they behave differently. Sixty percent is the sticker gap against Fable 5.1, which lists at $10 and $50. The 40% is a claim about what a job costs to run against Opus 5, and it comes from a line most buyers never read.

The Numbers

Anthropic's posted rates, per million tokens, as of Tuesday, Sept. 22, 2026.

Anthropic never posted sticker input and output rates for Opus 5, which is why the 40% run-cost claim cannot be checked against a price list.

The Discount Is Hiding In The Cache

The headline says 40% and the input line says 60%, and the two reconcile one row further down. Anthropic set cache reads for Claude Opus 5.5 at $0.20 per million tokens, writing that they are "60% less than Opus 5," and the company notes cache reads make up the majority of costs in agentic and coding workloads. Five-minute cache writes also drop, to $5 per million from $6.25.

That is the mechanism behind the 40%. A long-running coding agent that re-reads the same big pile of context hundreds of times a day is billed mostly on the cheap line, so the cache cut, not the input rate, is what moves its monthly bill. The discount is aimed at a workload, not at a price list.

Once a cheaper tier claims flagship-level output, the purchasing question moves from which model wins a leaderboard to what a finished task costs. A rival can answer that only by cutting its own prices.

The Flagship It Steps On Is Three Weeks Old

Claude Fable 5.1 arrived on Sept. 1, 2026, alongside the restricted-access Claude Mythos 5.1, priced at $10 and $50 per million tokens with cache reads cut to $0.25, a 75% reduction Anthropic said lowers typical workload costs by about 25% and highly agentic work by roughly 45%. Anthropic sold it hard, saying "Fable 5.1 sets a new standard for coding, knowledge work, and long-running problem-solving tasks," and quoted Craig Falls of Jane Street in the same announcement: "Solves more of our coding problems than Fable 5 or Opus 5."

Three weeks later the same lab says a cheaper model matches that on most work. The first thing Claude Opus 5.5 undercuts is the model Anthropic was selling in early September.

Read The Hedges

"Performs at the level of" and "on most work" are carrying serious weight, and neither phrase names the work that falls outside "most." Anthropic did attach numbers to the claim on Sept. 22, 2026:

What is missing is the comparison column. The same scores for Claude Fable 5.1 and Claude Opus 5 are absent, as is any independent run, so a reader can see the tests and still lack the one thing that would confirm the headline.

What settles this is the next few weeks. The rates are already public. Where Claude Opus 5.5 lands across tiers and surfaces, who gets access and when, and whether any competitor responds with a cheaper tier of its own are all open.

The Market That Priced The Release

Polymarket US, which bills itself as the first federally regulated prediction market in the United States, ran a market on exactly this event: "Next Claude Opus Released?", filed under Tech and Model Releases, with two rungs, by September 30, 2026 and by October 31, 2026. The chart on it opens September 11.

The day before the release the September rung was not a lock. It traded down to about 70% in the evening, climbed back through the night, sat in the mid-90s through Tuesday morning, and went to 100% around midday, when the model shipped.

Both rungs are settled by the release itself. A penny for No is what a market charges when the answer is already known, and the September rung is the one that decided it eight days early. The venue had not posted the formal resolution as of Tuesday afternoon, which is paperwork rather than doubt.

What the board could not tell anyone is the part that turned out to matter. It asked whether the next Opus would arrive, not what it would cost, and the price is the news.

About The Numbers

Anthropic has not posted input and output rates for Opus 5 alongside the new card, so the 40% run-cost claim cannot be checked against a sticker price here; the two Opus 5 figures that are public, cache reads and five-minute cache writes, are in the table above.

A price per million tokens is what Anthropic charges for text going into a model and text coming back out, which is why a job that re-reads the same large context all day is billed mostly on the cache read line rather than the headline input rate.

On a prediction market, the price is the chance: a No side quoted at 1 cent means the market puts the Yes answer at about 99%. Prices and chart readings for the release market are from Polymarket US on the afternoon of Tuesday, Sept. 22, 2026.

The benchmark figures are single published scores from Anthropic's own testing, each on one named test, and the 81.8% on OSWorld 2.0 is marked partial. The 1846 Elo is a rating on a comparison scale, so it carries meaning only against other models' ratings on that same scale, which are absent here. Trading volume on the release market is left out, since the venue's own page reported none for either rung.