Half off puts Sol at $2.50 in, $15 out per million, cache reads at $0.25. For agentic coding that last number is the one that matters — a long-horizon run re-reads the same context dozens of times, and cheap cache reads are what make a multi-step agent affordable rather than a treat.
What I’d do: point a throwaway coding task at it through OpenRouter, since the API is OpenAI-compatible and only the slug changes. Then check routing mode. OpenAI’s endpoint is the fast one at 3.59s latency and 46 tps; Azure and Bedrock post $5.00–$5.50 in and aren’t discounted, so a Balanced route can quietly hand you full price. Exacto if tool-calling accuracy matters. I’d confirm which provider served the request before believing the invoice.
The story — OpenRouter lists GPT-5.6 Sol, the flagship of OpenAI’s GPT-5.6 series, at 50% off: $2.50 per million input tokens and $15 output, down from $5.00 and $30.00, with cache reads at $0.25. It has a 1M context window, released Jul 9, 2026, with a Feb 2026 knowledge cutoff. Azure and Amazon Bedrock endpoints show no discount. Top traffic comes from Codex, Hermes Agent, and pi. (Source)