Changelog

The biggest update since launch. Reasoning v2 engine, Crucible Pro model tier, 256K context window, structured output mode, and a fully rebuilt dashboard.

Adaptive reasoning depth replaces fixed-depth search. Accuracy up 14 points on DOCLENS-v2. Latency down 22%. Automatic for all users.

Process full contracts, earnings reports, and regulatory filings in a single request. Context window doubled from 64K to 128K tokens.

Streaming via server-sent events is now supported across standard and fast reasoning modes. Deep mode streaming for the reasoning trace is on the roadmap for Q1 2026. See the Streaming page in the Core API section for implementation details and code examples.

The official Crucible SDKs for Node.js and Python have reached v1.0. Both SDKs are stable, fully typed, and cover the complete API surface including completions, streaming, tokenization, and webhook verification. Available via npm and PyPI.