What Is GPT-5.6?
GPT-5.6 is OpenAI’s frontier large language model family, the successor to GPT-5.5. It reached full general availability on July 9, 2026 across ChatGPT, Codex, and the OpenAI API, following a government-gated limited preview that opened June 26. Unlike prior generations, GPT-5.6 ships as three distinct tiers — Sol, Terra, and Luna — rather than a single model with a Pro variant, and is built primarily around agentic workflows: long-running, multi-step tasks completed with minimal human oversight, at a lower cost per task than GPT-5.5.
Key Features
- Three-tier lineup — Sol (flagship), Terra (value), and Luna (fastest/cheapest) all share the same ~1.05M-token context window and 128K max output; what differs is intelligence, speed, and price.
- Agentic-first design — headline gains are on multi-hour agent workloads, Terminal-Bench 2.1 coding, and long-horizon reasoning.
- Max and Ultra reasoning effort — Sol-specific settings; max gives the model more time to reason, ultra deploys subagents to accelerate complex work.
- Sol Fast — a premium option on Cerebras hardware running up to 750 tokens per second, at $12.50/$75 per million tokens.
- ChatGPT Work — launched alongside GPT-5.6, merging Codex into the ChatGPT desktop app with a 15-integration plugin directory, aimed at professional/agentic use.
- Multimodal input — text and image inputs across all three tiers.
Where You Can Use GPT-5.6
GPT-5.6 is live now in ChatGPT, the Codex coding environment, and the OpenAI API via the standard Responses and Chat Completions endpoints. It also powers the new ChatGPT Work product for agentic professional workflows.
GPT-5.6 Pricing (Confirmed)
Per million tokens:
- Sol — $5 input / $30 output
- Terra — $2.50 input / $15 output (GPT-5.5-class quality at roughly half the price)
- Luna — $1 input / $6 output
Prompt caching is supported across all tiers, with cached reads carrying roughly a 90% discount. For consumer access, GPT-5.6 is included in ChatGPT Plus, Pro, Business, and Enterprise plans.
Benchmarks
On Terminal-Bench 2.1, Sol Ultra scores 91.9% and base Sol 88.8%, ahead of GPT-5.5 (83.4%) and Claude Opus 4.8 (78.9%). On Agents’ Last Exam, Sol scored 53.6. All three tiers are rated High under OpenAI’s Preparedness Framework for cybersecurity and biological/chemical risk, without crossing the High threshold for AI self-improvement — the reason the initial rollout was staged rather than a standard public launch.
How GPT-5.6 Compares
GPT-5.6’s main rivals are Anthropic’s Claude Opus 4.8 and Google’s Gemini 3.5 Pro, which remains unreleased as of mid-July with its own date repeatedly slipping. Grok 4.5 undercuts GPT-5.6 sharply on price ($2/$6 vs Sol’s $5/$30) at somewhat lower benchmark scores. GPT-5.6 is strongest on agentic coding and Terminal-Bench performance; Claude leads on reasoning and writing quality. See our full GPT-5.6 vs Claude Opus 4.8 vs Gemini 3.5 Pro comparison for the detailed breakdown.
Is GPT-5.6 Worth It?
With GA now open to everyone, Terra is the pragmatic default for most production use — GPT-5.5-class output at roughly half the price. Sol is worth the premium for long-horizon agentic and coding work. Luna fits high-volume, cost-sensitive workloads. If you build or use AI agents, run coding pipelines, or care about cost-per-task at scale, GPT-5.6 is a strong default; if your priority is single-turn chat quality, Claude Opus 4.8 remains a strong alternative for reasoning and writing.
Last updated: July 17, 2026.