GPT-5.6.mdView source

OpenAI GPT-5.6 Family Integration

Historical July integration note. Current model ordering, supported settings and execution behavior are documented in the 2.6 technical record. Treat provider-performance, pricing and caching statements below as historical claims requiring current verification, not release guarantees.

Last Updated: 2026-07-10

On July 9, 2026, OpenAI released the GPT-5.6 series. OpenResponses provides comprehensive, day-one support for these models through the standard Responses API.

Model Tiers

The GPT-5.6 family abandons the “o” and “turbo” naming conventions for a tiered planetary naming scheme:

  1. gpt-5.6-sol (Flagship): The most capable model, optimized for complex reasoning, multi-step coding, and long-horizon agentic workflows. (The generic gpt-5.6 alias automatically routes to this model).
  2. gpt-5.6-terra (Balanced): A mid-tier model designed to provide a strict balance between intelligence and cost.
  3. gpt-5.6-luna (Cost-Efficient): The fastest, most cost-effective tier optimized for high-volume, lightweight tasks.

Core Capabilities

  • 1.05 Million Context Window: All models support a massive 1.05M token context.
  • 128K Output Limit: Can generate up to 128,000 output tokens in a single completion.
  • Image Handling at Scale: Models now natively accept images at their original dimensions (auto or original detail configurations).

New API Parameters & Controls

Persisted Reasoning

GPT-5.6 introduces expanded controls over internal thought processes.

  • reasoning_effort: Controls the target effort. The models now support none, low, medium, high, xhigh, and a new max option. max allows for the deepest possible internal thought process for long-horizon reasoning tasks.

Verbosity Control

  • verbosity: A new parameter that directly controls the output length independent of max_completion_tokens. Options are low (concise), medium (default), and high (comprehensive).

OpenResponses dynamically maps these parameters when these models are selected.

Explicit Prompt Caching Controls

Instead of opaque caching algorithms, GPT-5.6 supports explicit prompt caching with specific breakpoints. Caches have a guaranteed 30-minute minimum life, making repetitive tasks far more cost-effective.

Multi-Agent Orchestration (Beta)

Within the Responses API, GPT-5.6 natively handles routing between sub-agents and tool loops. (OpenResponses supports standard tool callbacks compatible with this).

Programmatic Tool Calling

Enhanced support for deterministic model-driven tool execution, allowing the model to chain multiple tools internally before returning the stream.

Application Support

In OpenResponses, GPT-5.6 models receive the highest priority sorting in the DynamicModelSelector. Because GPT-5.6 is optimized for long-horizon agentic workflows, the Computer Use tool is enabled for all models in the family (sol, terra, and luna).