Overview
GPT-6 Sol is the mid-tier GPT-6 model OpenAI shipped on September 22, positioned below the flagship GPT-6 Astra. It reuses Astra’s training methods with improvements to reasoning, factual reliability, coding, computer use and alignment. The headline is price: $2 per million input tokens and $10 per million output, half of what GPT-5.6 Sol charges at $4 and $20. That lands exactly on Claude Sonnet 5 pricing and half of Anthropic’s new Opus 5.5.
Key Features
- Priced per task: OpenAI’s launch material compares cost per completed task rather than token price alone, and Sam Altman says nothing on the market competes on that measure.
- Coding and computer use: 68.8% on the DeepSWE software engineering test, 60.5% on OSWorld computer use at extra-high effort and 56.4% on Agents’ Last Exam at maximum effort.
- Prompt cache improvements: Cached input reads now carry a 90% discount, developers can set explicit breakpoints for where a cached prefix ends, and changing reasoning effort or toggling tools no longer discards the cache.
- Agent workflow benchmark: AutomationBench spans 47 tools across sales, marketing, operations, support, finance and HR, where Sol scores 33.2% at extra-high effort at about $0.27 per task.
Use Cases
- Coding agents for the recurring work developers do: building features, reviewing code, debugging and analyzing data
- Business process automation across multi-step tool chains, where cost per task is the number that matters
- Computer use tasks that involve clicking, filling forms and moving through interfaces
- High-volume batch calls that need repeated model invocations without paying flagship prices at every step
Pros
- Prices are permanently halved, not an introductory promotion
- The caching changes cut the share of prompt tokens that need fresh processing by more than half in GitHub’s measurements, which matters for long conversations and agent loops
- Available in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu accounts
- The API name is gpt-6-sol, so migrating from the previous generation is a model string change
Pricing
The API charges $2 per million input tokens and $10 per million output, which OpenAI confirms is permanent pricing. Cached input reads carry a 90% discount and cache writes are billed at published rates. Sol is available in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu accounts. Check the official page for current prices and quotas.
Summary
GPT-6 Sol is the middle tier of the GPT-6 family. The claim that it comes close to Astra at half the price comes from OpenAI’s own cost accounting, and the independent read is more conservative: Artificial Analysis measured the Intelligence Index moving from 47 for GPT-5.6 Sol to 48, while average task cost fell from $1.99 to $1.06. The generational gain here is mostly in the invoice, not the leaderboard.
Version History
- OpenAI ships GPT-6 Sol and GPT-6 Luna at 50% below their GPT-5.6 counterparts (2026-09-22): OpenAI released GPT-6 Sol and GPT-6 Luna, filling in the mid-tier and light tier below GPT-6 Astra. Sol costs $2 per million input tokens and $10 per million output, Luna $0.10 and $0.50, both confirmed as permanent pricing rather than a promotion, 50% below the equivalent GPT-5.6 models. Both reuse Astra’s training methods, both ship in ChatGPT Work and Codex, and their API names are gpt-6-sol and gpt-6-luna. Prompt caching improvements landed the same day, raising the cached input discount to 90% and adding explicit cache breakpoints. Artificial Analysis measured clear cost efficiency gains, with task cost falling from $1.99 to $1.06 and the Intelligence Index moving from 47 to 48.