GPT-5.6 Sol is half price on OpenRouter and Vercel through Sept 18
It skips OpenAI's own API and hits only the two gateways used to gauge market share — $2.50/$15 per M, automatic, and analysts say that's the point.
Copy markdown
The cut, by the token
On OpenRouter and Vercel AI Gateway, Sol's default tier drops from $5/$30 to $2.50/$15 per million tokens; priority falls from $10/$60 to $5/$30. It applies automatically through Sept 18, cached reads and writes included — OpenAI's own API price is untouched.
It only lands where the scoreboard is
SemiAnalysis notes the discount covers just two gateways that are a tiny slice of OpenAI's tokens but the main public dataset for market-share estimates. Doubling Sol's volume there could read as a win over Anthropic that price, not quality, bought.
Why half price still pays
The logic is Jevons: TD Cowen found OpenAI's earlier 80% cut on Luna drove usage up 14x and revenue up 34%. Cheaper tokens grow the bill, so subsidizing a flagship on public rails can pay for itself.
What $2.50 actually buys
Sol keeps its 1M-token context and is tuned for command-line, multi-step, and long-horizon agentic coding. HN builders report canceling Claude subs after testing it — anecdotes, not benchmarks, but the price makes the trial cheap.
The whole tier is racing down
Sol isn't alone: Gemini 3.7 Flash is already half-off at $0.75/$3.75 per M through year-end, with DeepSeek and Kimi undercutting further. If you pin model versions, this is the window to re-benchmark your cost per task.