Eigen RadarAI
Archive
Analysis

Claude Opus 5.5 opens, and paid GitHub Copilot plans start receiving it

Anthropic opened Claude Opus 5.5 on 22 September and said that, when its own safeguards fire, some cyber tasks are finished on Opus 4.8 and some biology and frontier tasks on Opus 5. GitHub the same day began a gradual rollout of the model to paid Copilot plans, with a switch that lets business administrators turn it off.

Artificial Intelligence··Night
Two people sit with their backs to the camera at a sunlit desk, facing a large glowing glass console linked by thin light threads to a smaller older machine.

What Anthropic says it opened

Anthropic's own page presents Claude Opus 5.5 as open on 22 September. The account says a typical workload costs about 40 percent less than Opus 5, and that when the company's safeguards fire, cyber tasks are completed by Opus 4.8 while biology and frontier tasks are completed by Opus 5. Those are the company's statements about its own routing, not an outside measurement of either model. The same page says output is more than 30 percent faster and that cache reads are priced below the previous Opus. Anthropic can state a price and a speed claim without an independent lab repeating them. The development that both publishers share is narrower: the model is open, and a second company is putting it in front of paying developers.[1], [2]

Where GitHub is switching it on

GitHub's changelog says Claude Opus 5.5 opened on 22 September to Copilot Pro+, Max, Business and Enterprise users. The rollout is gradual. Business and Enterprise administrators can disable the model in policy settings. If the global default stays on, GitHub says new models turn on by themselves. GitHub also says the watermark on Copilot output does not change the meaning, the token count or the cost. That is a distribution note, not a second evaluation of Anthropic's speed or price claims. A team that does not want the model in the product has an administrator path; a team that leaves the default on receives it as the rollout reaches them.[2]

What a reader can treat as established

Two publishers, the model maker and the coding product, agree on the date and the name. Anthropic is the source for how its safeguards hand work to earlier models and for its own cost comparison. GitHub is the source for which paid plans are in the rollout and for the administrator switch. Neither page is an independent benchmark of the other.[1], [2]

References

  1. News sourceAnthropicClaude Opus 5.5 opens, and Anthropic completed some tasks on older models↩1↩2
  2. News sourceGitHub BlogGitHub Copilot opens Claude Opus 5.5 to paid plans, and admins can turn it off↩1↩2↩3
AI safety and regulationThe frontier model race

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →