GPT-5.6 Sol
OpenAI's most capable and security-hardened frontier model, in limited preview
Context window
Not announced
Input / 1M tokens
$5.00
Output / 1M tokens
$30.00
Provider
OpenAI
Preview pricing announced June 26, 2026; matches GPT-5.5 at $5/$30. GPT-5.6 introduces more predictable prompt caching with explicit cache breakpoints and a 30-minute minimum cache life; cache writes billed at 1.25x the uncached input rate, cache reads receive the 90% cached-input discount. Context window not stated in official preview materials. · Data verified 2026-07-05
GPT-5.6 Sol is the flagship tier of OpenAI's GPT-5.6 model family (Sol, Terra, Luna), previewed on June 26, 2026. Under the new naming scheme the number denotes the generation while Sol/Terra/Luna denote durable capability tiers that advance on their own cadence, giving developers clearer trade-offs across intelligence, speed, and cost. Sol is the highest-intelligence tier and advances the frontier on software engineering, computer use, professional knowledge work, scientific research, and cybersecurity. It adds two new modes: 'max' (a reasoning-effort setting giving the model the most time to think) and 'ultra' (subagent orchestration for complex, multi-step work). The launch is an unusual gated rollout coordinated with the U.S. government over security concerns rather than a normal public release. Context window and API model id are not stated in the official preview materials.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Limited preview only. Available to a small group of trusted partners and organizations via the API and Codex, with government case-by-case approval of customers during the preview period. General availability across ChatGPT and the API is planned 'in the coming weeks.' API model id is unconfirmed; 'gpt-5.6-sol' is an inferred placeholder pending official docs.
Strengths
- ✓Top-tier performance on agentic coding and terminal/computer-use benchmarks (Terminal-Bench 2.1: 88.8%, and 91.9% in Ultra mode)
- ✓New 'max' reasoning-effort mode for the deepest single-agent reasoning
- ✓New 'ultra' mode orchestrates subagents for complex, long-horizon tasks
- ✓Strong cybersecurity hardening and safety evaluations
- ✓Improved token efficiency versus GPT-5.5
Best for developers who...
When to choose it (and when not to)
Reach for GPT-5.6 Sol when...
- →You are an approved preview partner needing the most capable model for agentic software engineering
- →Complex multi-step tasks that benefit from subagent orchestration (Ultra) or maximum reasoning depth (max)
- →Frontier scientific research, professional knowledge work, or security-sensitive workloads
Look elsewhere if...
- ✕You need general public or unrestricted API access today (it is a gated, government-approved limited preview)
- ✕Cost-sensitive or high-volume workloads where Terra ($2.50/$15) or Luna ($1/$6) would suffice
- ✕You require a confirmed, documented context window or stable model id for production commitments
How to use it
- ›Use the 'max' reasoning-effort setting when a hard problem warrants the deepest single-agent thinking
- ›Reserve 'ultra' (subagent orchestration) for complex, decomposable, long-horizon tasks rather than simple queries
- ›Leverage explicit prompt-cache breakpoints to control caching and cost on repeated large prefixes
- ›Match the tier to the job: drop to Terra or Luna when Sol-level intelligence is not required
Quickstart
Pythonfrom openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-5.6-sol",
input="Refactor this module and add tests.",
reasoning={"effort": "max"},
)
print(response.output_text)Model id 'gpt-5.6-sol' is an inferred placeholder; confirm the exact id in OpenAI docs once GA. Access is limited to approved preview partners. The 'max' reasoning effort is a GPT-5.6 Sol feature.
API model id: gpt-5.6-sol
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Terminal-Bench 2.1 | 88.8% | GPT-5.6 Sol; Ultra mode (subagent orchestration) scores 91.9%, the highest in the family. |
| Terminal-Bench 2.1 (Ultra) | 91.9% | GPT-5.6 Sol Ultra, top score reported in the preview coverage. |
| ExploitBench | competitive with Mythos Preview | Using ~1/3 of output tokens |
Compare GPT-5.6 Sol with
GPT-5.6 Sol vs GPT-5.5
OpenAI - 1,050,000 tokens (128,000 max output) ctx
GPT-5.6 Sol vs GPT-5
OpenAI - 400,000 tokens (128,000 max output) ctx