Anthropic on Thursday released Opus 4.8, just 41 days after 4.7 went out - a cycle pace the company has not had since its start. The price stays at the level of the previous Opus, the model is available everywhere from day one. Translation: some pressure behind the scenes is real, and the 4.7 and 4.8 models are not ordinary incremental releases.

From inside, the reason is clear. Opus 4.7 was met with a cool reception from users; separately, OpenAI shipped a new update to Codex, and Google a new Gemini Flash. Anthropic could no longer keep Opus on the shelf for 41 days without an answer. Sonnet is three months old, Haiku seven - a standard cycle. Opus 4.8 in 41 days is a signal of blood in the water.

When you strip the marketing language out of the announcement, one thing is genuinely interesting: the new model is "more likely to mark uncertainties around its own work and less prone to making unfounded claims". Testers at Bridgewater (the fund, not the brand) say Opus 4.8 finds problems in input and output data that previous versions used to miss. Real-world translation: companies using it for analysis are finally getting a model that admits "I don't know" - instead of inventing a convincing answer.

Alongside Opus 4.8, Anthropic is also releasing Dynamic Workflows - a tool for managing hundreds of parallel "sub-agents" of a single large model. The technical translation of that: Claude Code can now carry out the migration of an entire project's code, from hundreds of thousands of lines, from start to merge, with the existing test set as the success metric. If this really works as they describe, it is what the industry calls the "agent economy" - the developer role is being renamed, not removed.

Hanging behind all this is also the unreleased card - the Mythos model, a class above Opus, whose pre-launch last month raised safety questions. Anthropic today signalled that the review is wrapping up: "in the coming weeks" Mythos should be available to all clients. Translation: the model is ready, they are just waiting for the safety layers to clear internal review. For the AI market, that is the biggest piece of news in the announcement - bigger than Opus 4.8.