GPT-5.6 Sol
Released Jul 9, 2026 · knowledge cutoff 2026-02
- Status
- Frontier
- Location
- United States
- Modality
- Multimodal
- Context window
- 1.1M
- Max output
- 128K
- Speed (medium effort)
- 64 tok/s · 6.09s to first answer token
- Price ($/MTok in / out)
- $5 / $30
- Cost per task (medium effort)
- $0.41
- Open weights
- No
Benchmarks7 of 8 reported
MMLU-Pro
GPQA Diamond
SWE-bench Verified
Terminal-Bench 2.1
Humanity's Last Exam
LMArena Elo
ARC-AGI-2
AIME (latest)
Strengths & weaknesses
Strengths
- Stays on one task for days without drifting or declaring premature victory
- Reaches the same answer in far fewer tokens than the model before it
- Security work — threat modelling, code review, blue-teaming — is a deliberate specialism
Weaknesses
- Still verbose for interactive chat, despite being efficient on agentic work
- Pre-deployment testing found it gaming its own evaluations, which makes any single score harder to trust
- Priced well above its Terra and Luna siblings, which close much of the gap on everyday work
The frontier tier of the three-model GPT-5.6 family (Sol / Terra / Luna), all released 2026-07-09. Context window and max output confirmed via OpenAI's developer docs. GPQA Diamond 94.6 and HLE 47.2 are OpenAI/Artificial Analysis-reported; MMLU-Pro 89.1 and ARC-AGI-2 92.5 (Max effort, per ARC Prize) are third-party. OpenAI published no SWE-bench or AIME figure for this family; SWE-bench Verified 96.2 is third-party, from Vals AI’s independent harness (leaderboard updated 2026-07-31, run at max reasoning effort), the same board that supplies Luna’s 93.0. AIME stays null. LMArena Elo 1485 is the "gpt-5.6-sol-xhigh" text-arena listing (arena.ai). Artificial Analysis Intelligence Index 59. Terminal-Bench 2.1 88.8% from the GPT-5.6 launch comparison table (Sol 88.8 / Terra 87.4 / Luna 84.7). Sol's price held at $5.00/$30.00 through OpenAI's 2026-07-30 cuts to the other two tiers, which also added an optional fast mode: up to 2.5x the speed at twice the price.
For developers
API model strings
- openai
gpt-5.6-sol
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor
News
In the news
- Jul 30, 2026
OpenAI cuts GPT-5.6 Luna by 80% and Terra by 20% three weeks after launch
- Jul 29, 2026
OpenAI opens a free ChatGPT tier for 100,000 academic researchers
- Jul 29, 2026
OpenAI updates its Hugging Face breach disclosure: an Artifactory zero-day and four compromised accounts
- Jul 24, 2026
ARC Prize verifies Claude Opus 5 at 30.2% on ARC-AGI-3, quadrupling the previous best
- Jul 24, 2026
Artificial Analysis puts Claude Opus 5 first on its Intelligence Index
- Jul 23, 2026
Frontier-Bench v0.1 launches as the successor to Terminal-Bench
- Jul 21, 2026
OpenAI discloses that its models escaped an eval sandbox and breached Hugging Face
- Jul 17, 2026
Artificial Analysis puts Kimi K3 third on its Intelligence Index
- Jul 10, 2026
OpenAI credits GPT-5.6 Sol Ultra with a proof of the 50-year-old cycle double cover conjecture
- Jul 9, 2026
OpenAI ships GPT-5.6 (Sol, Terra, Luna) to the public
- Jun 26, 2026
OpenAI previews GPT-5.6 Sol, Terra and Luna under government-gated release