OpenAI cuts GPT-5.6 Sol to $4/$20 — undercutting Claude Opus 5
The 20%/33% cut runs three months on API and Codex credits only; watch the 272K-token cliff, and OpenRouter's own Sol promo may still be cheaper.
Copy markdown
Input down 20%, output down 33%
GPT-5.6 Sol drops from $5/$30 to $4/$20 per million tokens, with cached input at $0.40 — the second cut on the 5.6 family in a month. At $20 output it now undercuts Claude Opus 5's $25, so recompute your cost estimates.
API and Codex credits only — subscriptions unchanged
The discount reaches the API and eligible ChatGPT Work/Codex credit plans; standard subscription pricing stays put. It is promotional, holding at least through November 21, 2026 — a window worth planning batch jobs around.
Mind the 272K-token cliff
The $4/$20 rate only applies under 272K input tokens. Cross that line and you pay 2x input and 1.5x output, so the discount evaporates on big-context runs — keep prompts under the threshold.
OpenRouter may still be cheaper
OpenRouter is running a separate 50% promo on Sol routes, so the lowest price to the same model might not be OpenAI direct. Check both before you re-point your API base URL.
Elsewhere: DeepSeek ships a 1M-context vision Flash
DeepSeek V4 Flash Vision Exp hit OpenRouter on Aug 21 at $0.22/$0.66 per million with a 1M-token context — a 13B-active MoE tuned for document and chart understanding and multimodal agent loops.