The AI Post
Agents & CodingOpen ModelsEnterpriseFundraisingGenerative MediaGovernanceInferenceInfrastructureLegal & SafetySector Impact
← Front Page Model Releases · Anthropic

Opus 5.5 is said to arrive Tuesday, 20 percent cheaper than Opus 5

Accounts circulating since Sunday put the model at $4 and $20 per million tokens, testing under the codename claude-wafer-eap and strongest at 3D and front-end work.

Anthropic will release Claude Opus 5.5 as early as Tuesday at prices a fifth below its current flagship, according to accounts that began circulating on Sunday. They put the model at $4 per million input tokens and $20 per million output, with cache reads at $0.20 and writes at $5. They put it in testing under the codename claude-wafer-eap.

The capability claims place it alongside GPT-6 Astra and Fable 5.1, ahead of both on some work. Accounts single out 3D generation and front-end development, and describe shorter, cleaner output and a better grip on fine detail. One developer says it produced a Waymo vehicle in 3D with the badge and roof lidar modelled. They describe its thinking as always adaptive.

The version number moved. Several accounts say the model began as Opus 5.2 and took the 5.5 name late, because the changes outgrew a point release. Anthropic ran a brief Opus 5.2 test between 14 and 17 September, routing a slice of Opus 5 traffic to an unreleased checkpoint. One account has the context window falling from 936,000 tokens to 872,000.

The pricing would be a real cut. Anthropic lists Opus 5 at $5 per million input tokens and $25 per million output, with cache reads at $0.50 and writes at $6.25. The leaked pair is exactly 20 percent below both, and the cache read would fall by 60 percent. For long-context agent work, the cache line matters more than the headline rate.

Anthropic has confirmed none of it. No Opus 5.5 appears on the price list, and the company has published no model card and no API identifier. Reuters reported on Friday that the company is weighing a new model to answer GPT-6 Astra, days after Dario Amodei urged the industry to slow down. Polymarket puts the next Opus at 72 percent to land by Thursday.

The account that carried the claim furthest, kimmonismus, relaying a leaker called Lyra, said on Monday that he cannot confirm it. Partners got 24 hours with the model, one account says. GPT-6 Sol is due the same week. A published price list would settle the numbers within a day.

Sources 15 sources

  1. Commentary @kimmonismuspost on X
  2. Commentary @kimmonismuspost on X
  3. Commentary @kimmonismuspost on X
  4. Primary Anthropic — published API price list
  5. Primary @AnthropicAIpost on X
  6. Press TestingCatalogpost on X
  7. Press Polymarketpost on X
  8. Press Polymarket — citing Reuterspost on X
  9. Commentary OrcaRouter — analysis of the leak's sourcing
  10. Press Pasquale Pillitteri
  11. Press WinCentral
  12. Commentary @CaptainInsightXpost on X
  13. Commentary @AIJonHupost on X
  14. Commentary @yulmu_coffeepost on X
  15. Commentary @tom_knockkpost on X