OpenAI Launches Ultrafast Mode for GPT-6.1 Sol
OpenAI launches Ultrafast mode for GPT-6.1 Sol across API, Codex, and ChatGPT Work, charging 6x standard rates for up to eight times faster inference.
OpenAI released its Ultrafast service tier for GPT-6.1 Sol on October 8, 2026, bringing higher throughput to its API, the Codex development platform, and ChatGPT Work. The company claims the tier pairs reasoning capabilities near the flagship GPT-6 Astra with generation speeds up to eight times faster than standard Sol. But that speed bump carries steep financial trade-offs.
The Price of Rapid Inference
On the API, GPT-6.1 Sol Ultrafast costs $12.00 per million input tokens and $60.00 per million output tokens. That is exactly six times the base price of Sol Standard ($2.00 input and $10.00 output) and roughly 20 percent higher than the standard rate for GPT-6 Astra. Developers requesting strict regional data residence in the United States or the European Union pay an additional 10 percent surcharge.
Interface access is equally narrow. Only subscribers on the $500 monthly Pro plan, enterprise accounts on usage-based contracts, and eligible Edu accounts can enable Ultrafast in Codex and ChatGPT Work. Subscribers on Plus, $100 Pro, $200 Pro, and Business tiers cannot access the feature, and OpenAI has shared no timetable for expanding availability. On the $500 Pro plan, queries run in Ultrafast consume workspace quotas eight times faster than normal. Once that limit runs out, overage fees kick in at the 6x multiplier.
Targeted at Agents, Constrained by Network Latency
OpenAI built Ultrafast primarily for automated agent workflows. Systems that write code or browse the web run dozens of reasoning steps in sequence. When each intermediate step takes seconds to return, an entire autonomous loop drags to a crawl.
Raw model acceleration only solves half the problem. In its official developer documentation, OpenAI notes that repeated TCP and TLS handshakes can wipe out these hardware gains during rapid tool-calling loops. The company advises engineering teams to use persistent WebSocket connections instead of firing off isolated HTTP requests. In code, enabling the mode requires setting service_tier: "ultrafast" alongside model: "gpt-6.1-sol".
Company Benchmarks and Unconfirmed Timelines
The claim of generation speeds up to eight times faster comes straight from OpenAI internal testing. Independent benchmarking groups like Artificial Analysis have not published audited measurements for time-to-first-token or sustained generation throughput under real production traffic. The 8x figure remains a supplier benchmark rather than an independently verified metric.
OpenAI has also shared no timeline for bringing GPT-6.1 Sol or Ultrafast to the standard consumer ChatGPT interface. For now, Ultrafast serves as specialized infrastructure for teams willing to absorb higher infrastructure bills to shave seconds off complex agent loops.