Crucible 3.0 is the largest update we have shipped since launch. This release touches the core model, the API surface, and the developer experience. Here is everything that is new.
Reasoning v2 Engine
The reasoning engine that powers Deep Mode has been rebuilt from the ground up. Reasoning v2 improves multi-document synthesis accuracy by 14 points on our internal benchmarks and cuts median latency in Deep Mode by 22%. Existing integrations require no changes. The upgrade is automatic for all users.
New Model: Crucible Pro
Crucible 3.0 introduces a new model tier: Crucible Pro. Pro is a larger, more capable model designed for the most demanding reasoning tasks — complex legal analysis, multi-source financial research, and high-stakes document review where accuracy is non-negotiable.
Crucible Pro is available on all paid plans. See the Models documentation for a full breakdown of capabilities and pricing.
Expanded Context Window
Crucible-1 now supports a 256K token context window, up from 128K. This covers the full length of most financial filings, legal agreements, and technical specifications without chunking.
Structured Output Mode
You can now pass a JSON schema directly in your request and Crucible will guarantee that the output matches the schema. No prompt engineering required. Schema validation is enforced at the generation level, not post-processed.
Dashboard Redesign
The dashboard has been rebuilt with new usage analytics, per-key rate limit controls, and a live request log. You can now inspect recent requests and their reasoning traces directly from the dashboard without making API calls.
Crucible 3.0 is the largest update we have shipped since launch. This release touches the core model, the API surface, and the developer experience. Here is everything that is new.
Reasoning v2 Engine
The reasoning engine that powers Deep Mode has been rebuilt from the ground up. Reasoning v2 improves multi-document synthesis accuracy by 14 points on our internal benchmarks and cuts median latency in Deep Mode by 22%. Existing integrations require no changes. The upgrade is automatic for all users.
New Model: Crucible Pro
Crucible 3.0 introduces a new model tier: Crucible Pro. Pro is a larger, more capable model designed for the most demanding reasoning tasks — complex legal analysis, multi-source financial research, and high-stakes document review where accuracy is non-negotiable.
Crucible Pro is available on all paid plans. See the Models documentation for a full breakdown of capabilities and pricing.
Expanded Context Window
Crucible-1 now supports a 256K token context window, up from 128K. This covers the full length of most financial filings, legal agreements, and technical specifications without chunking.
Structured Output Mode
You can now pass a JSON schema directly in your request and Crucible will guarantee that the output matches the schema. No prompt engineering required. Schema validation is enforced at the generation level, not post-processed.
Dashboard Redesign
The dashboard has been rebuilt with new usage analytics, per-key rate limit controls, and a live request log. You can now inspect recent requests and their reasoning traces directly from the dashboard without making API calls.
Crucible 3.0 is the largest update we have shipped since launch. This release touches the core model, the API surface, and the developer experience. Here is everything that is new.
Reasoning v2 Engine
The reasoning engine that powers Deep Mode has been rebuilt from the ground up. Reasoning v2 improves multi-document synthesis accuracy by 14 points on our internal benchmarks and cuts median latency in Deep Mode by 22%. Existing integrations require no changes. The upgrade is automatic for all users.
New Model: Crucible Pro
Crucible 3.0 introduces a new model tier: Crucible Pro. Pro is a larger, more capable model designed for the most demanding reasoning tasks — complex legal analysis, multi-source financial research, and high-stakes document review where accuracy is non-negotiable.
Crucible Pro is available on all paid plans. See the Models documentation for a full breakdown of capabilities and pricing.
Expanded Context Window
Crucible-1 now supports a 256K token context window, up from 128K. This covers the full length of most financial filings, legal agreements, and technical specifications without chunking.
Structured Output Mode
You can now pass a JSON schema directly in your request and Crucible will guarantee that the output matches the schema. No prompt engineering required. Schema validation is enforced at the generation level, not post-processed.
Dashboard Redesign
The dashboard has been rebuilt with new usage analytics, per-key rate limit controls, and a live request log. You can now inspect recent requests and their reasoning traces directly from the dashboard without making API calls.