OpenAI released the GPT-5.6 family on July 9, 2026, marking a shift from one general-purpose model to three distinct tiers. GPT-5.6 Sol is the flagship, GPT-5.6 Terra handles everyday work, and GPT-5.6 Luna focuses on speed and cost efficiency.
GPT-5.6 Sol: The Flagship
Sol is OpenAI’s strongest model to date. On the Agents’ Last Exam, an evaluation covering 55 professional fields, GPT-5.6 Sol scored 53.6%, beating Claude Fable 5’s 40.5% by 13.1 points. The gap widens when considering cost: even at medium reasoning, Sol beats Fable 5 while costing roughly one-quarter as much.
For coding, Sol set a new high on the Artificial Analysis Coding Agent Index with a score of 80, surpassing Fable 5 by 2.8 points while using less than half the output tokens and costing about one-third less.
A new feature called ultra mode coordinates four agents in parallel across multiple workstreams, trading higher token usage for faster results on demanding tasks. On BrowseComp, SEC-Bench Pro, and Terminal-Bench 2.1, the multi-agent approach consistently outperformed single-agent configurations.
Terra and Luna: Same Family, Lower Cost
Terra targets everyday work and performs just above Fable 5 on coding benchmarks. Luna, the cheapest option, outperforms Claude Opus 4.8 while running in roughly one-third the time. Both models use fewer output tokens than competitors, directly translating to lower API bills.
OpenAI positions this tiered approach as a response to how developers actually use AI: most tasks don’t need flagship-level compute, and paying for unnecessary power wastes money.
New API Features
Programmatic Tool Calling in the Responses API lets GPT-5.6 write and run lightweight programs that coordinate tools, filter intermediate data, and adapt workflows on the fly. Instead of passing every tool response back through the model, programs handle routine steps internally.
max reasoning gives the model more time to explore alternatives. ultra goes further with parallel agent coordination by default.
Benchmarks and Pricing
On the Artificial Analysis Intelligence Index, Sol with max reasoning falls within one point of Fable 5 while completing tasks in 61% less time at roughly half the estimated cost. The model also sets state-of-the-art results on Terminal-Bench 2.1 and DeepSWE, both testing real-world engineering workflows in actual codebases.
All three GPT-5.6 models launched across ChatGPT, Codex, and the OpenAI API on July 9, 2026. OpenAI has not disclosed exact per-token pricing for the three tiers, but Terra and Luna should be significantly cheaper than GPT-5.5 and competing frontier models.
What the Three-Model Split Means
The structure mirrors how cloud providers offer different instance sizes: developers match compute to task complexity rather than using a one-size-fits-all approach. The introduction of ultra mode with parallel agents also signals where AI is heading: coordinated multi-agent systems rather than single-model inference.
Frequently Asked Questions
What are the differences between GPT-5.6 Sol, Terra, and Luna?
Sol is the most capable model for complex reasoning and coding. Terra balances performance and cost for everyday tasks. Luna prioritizes speed and affordability, outperforming Claude Opus 4.8 at a fraction of the cost.
How much does GPT-5.6 cost compared to GPT-5.5?
OpenAI has not published exact pricing yet, but benchmark data shows Terra and Luna achieve comparable or better results than GPT-5.5 at around one-sixteenth the cost. Sol delivers state-of-the-art performance at roughly half the estimated cost of competing frontier models.
What is the ultra mode in GPT-5.6?
Ultra coordinates four agents in parallel by default, splitting complex tasks across multiple workstreams. It trades higher token usage for faster completion times and stronger results on demanding evaluations.
Can I use GPT-5.6 in the API right now?
Yes, all three GPT-5.6 models are available through the OpenAI API as of July 9, 2026. Some ChatGPT accounts may not see Sol in the model picker immediately; OpenAI is rolling out access gradually.
How does GPT-5.6 compare to Claude Fable 5?
Sol beats Fable 5 by 13.1 points on Agents’ Last Exam and 2.8 points on the Coding Agent Index. Even at medium reasoning, Sol outperforms Fable 5 at roughly one-quarter the estimated cost.
