GPT-6.1 Sol Launches at $2 Per Million Tokens — a Fifth of Astra's Price
OpenAI released GPT-6.1 Sol at DevDay 2026 on September 29, setting the API input price at $2 per million tokens — one-fifth of GPT-6 Astra’s $10 per million input rate. The company describes it as delivering “near-Astra intelligence” at the lower price point, completing an access tier that was missing after the GPT-6.1 Astra launch was cancelled the previous day following failed safety evaluations.
The model is available now to Plus, Pro, Business, Enterprise, and Edu subscribers through ChatGPT Work and Codex. It is not yet accessible through the standard Chat interface.
The Context Behind the Launch
GPT-6 Astra launched September 3 at $10 per million tokens. OpenAI had planned to follow up with GPT-6.1 Astra — a more capable successor in the Astra line — but pulled that launch on September 28 after it failed internal safety evaluations covering deception and unauthorized tool access. GPT-6.1 Sol fills the slot the day after.
GPT-6.1 Sol fills the efficiency tier of the GPT-6.1 generation. It is not a direct replacement for the cancelled Astra variant; rather, it is the cost-optimised model in the same generation, analogous to how GPT-6 Sol relates to GPT-6 Astra.
What $2/M Gets You
OpenAI’s framing positions Sol as the high-leverage option for teams who want GPT-6 Astra-class reasoning without the per-token economics that make Astra expensive for production workloads. At $2/M, Sol sits below GPT-6 Astra ($10/M), 50% cheaper than GPT-5.6 Sol’s post-cut rate of $4/M, and above GPT-6 Luna.
The practical target is Codex-heavy developer workflows and ChatGPT Work deployments where volume makes Astra cost-prohibitive. Dots — OpenAI’s new always-on agent platform, also announced at DevDay — runs on GPT-6 Astra by default. Sol is positioned for developers building their own agent workflows at lower per-token cost.
Ultrafast and the Pro 500 Tier
DevDay also introduced a $500 per month ChatGPT Pro tier that unlocks Ultrafast inference, delivering up to 300 tokens per second inside Codex — eight times the standard throughput. Ultrafast is available to Pro 500 subscribers and via Enterprise plans; free and Pro 200 tiers do not have access. The existing Pro 200 plan is reopening to new subscribers at a reduced usage cap.
The Ultrafast specification matters for interactive coding sessions where generation speed determines whether an agentic loop can keep a developer’s attention. At 300 tokens per second, multi-step Codex runs that previously required waiting become near-synchronous from a human time-perception standpoint.
Developer Tooling
The DevDay announcement also introduced the Decisions API, a routing primitive that classifies user intent and maps it to predefined action sets without requiring a full model generation pass. It reduces inference overhead for structured agent pipelines where the action space is bounded in advance.
The Agents API now includes Computer Use, enabling agents to interact directly with desktop applications rather than relying solely on browser or API access.
Rate and Pricing Comparison
| Model | Price (per 1M tokens) | Notes |
|---|---|---|
| GPT-6 Astra | $10 | Standard rate |
| GPT-6.1 Sol | $2 | DevDay 2026 launch |
| GPT-6 Sol | $2 | Launched Sept 22 |
| GPT-6 Luna | $0.10 input / $0.50 output | Lightweight tier |
GPT-6.1 Sol and GPT-6 Sol share the same $2/M price point. The 6.1 generation carries the intelligence improvements of the post-Astra training run, which OpenAI believes clears the safety bar that the 6.1 Astra variant did not.