Proprietary   OpenAI · Released July 9, 2026 (General Availability)

GPT-5.6 Sol

OpenAI's Three-Tier Frontier Family — Sol, Terra & Luna

Architecture Proprietary transformer (details undisclosed)
Parameters Undisclosed (frontier proprietary)
Context Window 1 million tokens (all three tiers)
Modalities Text, Code, Images, Audio
Input Pricing Sol: $2.50* / Terra: $2.50 / Luna: $1.00 (*estimated)
Output Pricing Sol: $15.00* / Terra: $15.00 / Luna: $6.00 (*estimated)
Open Weights No — proprietary, available via API, ChatGPT, Codex, Microsoft 365 Copilot
Availability ChatGPT, OpenAI API, Codex, Microsoft 365 Copilot. Cerebras deployment for Sol at 750 tok/s.
Why This Model Matters The GPT-5.6 family represents OpenAI's strategic shift from single-model releases to a three-tier architecture where Sol, Terra, and Luna are durable capability tiers that can advance independently on their own cadence. Sol is the flagship — achieving 91.9% on TerminalBench 2.1, the highest recorded score, and powering the most demanding agentic and reasoning workloads. Terra delivers GPT-5.5-level performance at roughly half the cost, making it the natural upgrade for existing GPT-5.5 deployments. Luna achieves an impressive 82.5% on TerminalBench at just $1/MTok input — redefining what's possible at the budget tier. All three models share a 1M-token context window. The rollout was unprecedented: a limited preview on June 26 accessible to approximately 20 government-approved partner organizations, marking the first frontier model launch coordinated with government oversight, followed by general availability on July 9. OpenAI also announced GPT-5.6 Sol deployment on Cerebras wafer-scale infrastructure, achieving up to 750 tokens per second — roughly 10-15x typical frontier model API speeds.
Visit OpenAI → ← All Releases
Performance

Benchmarks & Evaluation Results

TerminalBench 2.1 (Sol Ultra)

91.9% — highest score recorded

TerminalBench 2.1 (Terra)

Matches GPT-5.5 at roughly half the price

TerminalBench 2.1 (Luna)

82.5% at just $1/MTok input

SWE-bench Verified

95% (Sol)

Artificial Analysis Intelligence Index

#1 overall (Sol)

In Depth

Full Analysis of GPT-5.6 Sol

The GPT-5.6 family launch was the most consequential OpenAI release since GPT-4. The three-tier architecture — Sol (flagship), Terra (balanced), and Luna (efficient) — gives developers clear upgrade paths within OpenAI's ecosystem. Sol achieves a record 91.9% on TerminalBench 2.1, a comprehensive agentic coding and system operations benchmark, making it the most capable model for autonomous software engineering tasks. Terra matches GPT-5.5's performance across most evaluations while costing

roughly half as much — a direct response to pricing pressure from DeepSeek and Moonshot AI. Luna, at just $1/MTok input, is OpenAI's most aggressive pricing move yet, targeting the high-volume, cost-sensitive tier where open-source models have been gaining traction. All three models share architectural improvements including enhanced reasoning chains, improved instruction following, and the 1M-token context window. The Cerebras deployment for Sol — achieving 750 tokens per second — changes

the user experience for interactive agentic workflows, making real-time model interaction feel instantaneous. The GPT-5.6 family is now the preferred model in Microsoft 365 Copilot and powers Codex, OpenAI's next-generation coding assistant.

Assessment

Strengths & Considerations

Strengths

  • + Sol achieves record 91.9% on TerminalBench 2.1 — highest ever recorded
  • + Three-tier pricing enables cost optimization across workloads
  • + 1M-token context across all tiers — not just the flagship
  • + Cerebras deployment delivers 750 tok/s — 10-15x typical frontier speeds
  • + Deep Microsoft 365 Copilot and Azure ecosystem integration
  • + Government-coordinated safety framework provides regulatory clarity
  • + Most extensive third-party ecosystem of any model family

Considerations

  • Proprietary with no self-hosting or weight access
  • Sol pricing premium estimated at $2.50/$15 per MTok
  • Limited availability at GA — global rollout continues through July
  • No architectural transparency — training details remain undisclosed
  • Government coordination may raise concerns about access equity
Applications

Best Use Cases for GPT-5.6 Sol

Where this model excels and the types of workloads it is best suited for.

Use Case 1

Most demanding reasoning and multi-step coding tasks (Sol)

Use Case 2

Production deployments at GPT-5.5 quality with lower cost (Terra)

Use Case 3

High-volume, cost-sensitive applications (Luna)

Use Case 4

Enterprise deployments via Microsoft 365 Copilot

Use Case 5

Autonomous agentic workflows requiring maximum capability

Use Case 6

Research and analysis requiring frontier intelligence with regulatory clarity

Getting Started

How to Use GPT-5.6 Sol

Quick Start Guide

Access GPT-5.6 models through the OpenAI API at platform.openai.com — select the appropriate model (gpt-5.6-sol, gpt-5.6-terra, or gpt-5.6-luna). ChatGPT Plus and Pro subscribers get Sol access through ChatGPT. Microsoft 365 Copilot users receive GPT-5.6 as the preferred model automatically. For Cerebras-speed Sol inference, contact OpenAI for dedicated access.

Visit OpenAI → ← All July 2026 Releases