September 23, 2026
GPT-6 Sol is half the price of Opus 5.5: $2 per million input tokens
On September 22, GPT-6 Sol launched at $2 per 1 million input tokens, twice as low as Claude Opus 5.5. Output costs $10 versus $20.
Dan Shipper
@danshipper
GPT-6 Sol and Opus 5.5 launched today. We at @every have extensively tested both models on real work. Here is our head-to-head vibe check:
· 16.1K views
Opus 5 cost $5 for input and $25 for output per million tokens. Anthropic cut Opus 5.5 pricing to $4/$20, cached tokens to $0,20, and claimed generation more than 30% faster. GPT-6 Sol costs $2/$10, so the same API budget buys twice as much input and output as Opus 5.5.
Access and settings. GPT-6 Sol already works in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise, and Edu plans. In the API, the model is called as `gpt-6-sol`, lets you choose `reasoning.effort` from `none` to `max`, and OpenAI recommends the Responses API for built-in tools and function calling.
Project migration. In Claude Code, `/claude-api migrate this project to claude-opus-5-5` changes the model ID and parameters, then provides a checklist for manual verification. In the Claude API, replace `claude-opus-5` with `claude-opus-5-5`: thinking is now always enabled, its depth is set with `output_config.effort`, and legacy thinking values and forced `tool_choice` types `any` and `tool` return 400.
Testing it at work. In an Every test, Kieran Claassen got 14 lines of Ruby from Opus 5.5 instead of 26 from Fable 5.1. His run showed 427 requests per second and met 17 of 20 latency targets, while he rated its coding ability at 90% of Fable. Tyler Nishida asked Opus 5.5 to make a golf game with one prompt: the model launched its own team of agents and worked for 1 hour 52 minutes until access ran out.
GPT-6 Sol has a context window of 1 050 000 tokens versus 1 million for Opus 5.5, and both models produce up to 128 thousand tokens.
