The AI startup is stealth testing a new checkpoint under the codename `claude-wafer-eap` for its next major model update. This version, known as Opus 5.5, mandates adaptive thinking with no "off" mode and retires the forced tool use functionality. Partners received 24 hour access to the system prior to a wider release scheduled for Tuesday, though the rollout could begin as early as Monday.

Pricing for the new model is $4 per Mtok for input and $20 per Mtok for output. Cache read and write costs are $0.2 per Mtok and $5 per Mtok respectively. These technical specifications come from leaked data indicating the model will compete with future high-end releases from OpenAI, specifically GPT-6.