GPT-5.6 Sol

GPT-5.6 Sol is the top tier of OpenAI's mid-2026 frontier line, above Terra and Luna. All three tiers share the same shape — a 1,050,000-token context window, a 128,000-token output ceiling and a 16 February 2026 knowledge cutoff — so what separates Sol is capability and price, not capacity. Input is text and images; output is text. Its strongest showing is in agentic coding. Running with maximum reasoning effort inside OpenAI's own Codex harness, Sol set a new high on the Artificial Analysis Coding Agent Index at 80, about 2.8 points above Claude Fable 5, and did so while spending under half the output tokens and under half the wall-clock time. On the broader Artificial Analysis Intelligence Index it finishes a close second to Fable 5 at roughly a third of the cost. The exception is telling. On SWE-Bench Pro, which measures resolving real GitHub issues in their original repositories rather than in a curated harness, GPT-5.6 trails Anthropic's models by a clear margin. The gap between Sol's index-topping harness scores and its weaker showing on unmodified real-world repositories is the most useful thing to know about the model before adopting it. Pricing is USD 5 per million input tokens and USD 30 per million output tokens, with cached input at USD 0.50. There is a discontinuity worth planning around: any request whose input exceeds 272,000 tokens is billed at twice the input rate and one and a half times the output rate for the entire request, not just the excess. A million-token context is therefore available, but the last three quarters of it cost materially more than the first quarter. Reasoning tokens are billed as output. Sol is served through the OpenAI API and ChatGPT, with tool support covering function calling, web search, file search, code interpreter, computer use and image generation, plus structured outputs, streaming and prompt caching.

▸Videos

No videos yet.

▸Reviews