Anthropic appears to be close to releasing Claude Opus 5, and now signals point to internally tested into the hands of a limited partner. The most striking artifact so far is a research entry that briefly appeared in the coding tool’s model selector earlier this month and then was deleted within a few hours. It was listed as an early access preview with step-by-step controls, security fallbacks, a contextual window of one million tokens, and setup that required some extra effort. The corresponding model string was then listed in Google’s Vertex quota catalog, repeating the sequence that preceded the Sonnet 5. There is no pricing, system map or public model ID yet available, so the spec list remains preliminary.
Early preparations for Claude Opus 5 have been spotted again. Seems inevitable.
— M1 (@M1Astra) July 23, 2026
If this happens, the model will appear in the Claude app model selector for paid tiers, Claude Code, the Claude platform, and three cloud partners, likely replacing Opus 4.8 rather than running alongside it. The digression detail is quite telling: the routing of marked clues up to version 4.8 is similar to the pattern used above the Opus level, hinting that this release is positioned closer to the boundary than its number suggests.
This positioning matters because Opus is under pressure from both sides. Sonnet 5 came out in late June with a rating close to 4.8 at a fraction of the cost, while Fable and Mythos are above that. Subscription access to Fable 5 ended on July 19th with usage credits. A stronger Opus would allow Max, Team and Enterprise customers to move somewhere without paying Mythos rates, which is the practical reason why talk about the timing has taken hold this week.
Cadence supports this assumption without proving it:
- 70 days from 4.6 to 4.7
- 42 from 4.7 to 4.8
- 56 since then
Expectations are aimed at clear progress in coding work, although the reception of Sonnet 5 is a reminder that they are not always met. Anthropic said nothing.