GPT-5.6 Sol
OpenAI · Released Jul 9, 2026
$4
Input price / MTok
#25 of 33$20
Output price / MTok
#25 of 331.1M
Context window
#5 of 3478.5
ExploitBench (cybersecurity)
#2 of 276.9
ScreenSpot-Pro (desktop automation)
#2 of 2Summary
GPT-5.6 Sol is a proprietary language model from OpenAI, released on Jul 9, 2026.
At $4 input and $20 output per million tokens, it ranks #25 of 33 for input price, placing it mid-range on input cost.
Its context window of 1.1M tokens ranks #5 of 34 for context depth in this dataset.
It targets coding, agents, reasoning workloads.
Flagship of OpenAI's GPT-5.6 family (Sol/Terra/Luna tiers), announced as a limited preview Jun 26 2026 with general availability Jul 9 2026. Strong on long-horizon, command-line, and multi-step coding tasks. On the independent ARC-AGI-3 public set, OpenAI reported the official-harness score understates it: enabling retained reasoning and compaction lifted the result from 13.3% to 38.3%, beating Claude Opus 5's 30.2% on the same set — a real, reproducible optimization rather than a benchmark discrepancy, but flagged here since the two settings are not on by default. API identifier gpt-5.6-sol, knowledge cutoff Feb 16 2026. Superseded as OpenAI's flagship by GPT-6 Astra (Sep 3 2026).
Benchmarks
| Benchmark | Score | Max | % | Rank | Source |
|---|---|---|---|---|---|
| ExploitBench (cybersecurity) | 78.5 | 100 | 78.5% | #2/2 | OpenAI (cited as the prior-generation co... |
| ScreenSpot-Pro (desktop automation) | 76.9 | 100 | 76.9% | #2/2 | OpenAI (cited as the prior-generation co... |
Similar Models
Compare all 2 modelsModels with the most shared benchmark coverage and closest scores to GPT-5.6 Sol.
Compare GPT-5.6 Sol side by side
Pre-filled with 1 similar model