GPT-6.1 Sol Ultrafast: Access, Pricing and API

GPT-6.1 Sol Ultrafast became available on October 8, 2026 in Codex, ChatGPT Work and the Responses API. The product route requires Pro $500 or an eligible Enterprise or Edu agreement; the API route is available to API users with separate token billing. This turns September’s promised speed tier into a usable option. OpenAI’s dated changelog establishes the release, while the current documentation explains the access and cost boundaries.

The September 29 Sol launch introduced the model and described Ultrafast as forthcoming. The October 8 change is the availability of that tier, not a replacement model launch. Keep the exact model ID, gpt-6.1-sol, when comparing configurations. The Sol reference covers the underlying model; this report covers the new service option and its practical rollout.

GPT-6.1 Sol Ultrafast access and workspace eligibility

Pro $500 can use Ultrafast in Work and Codex. Other self-serve plans cannot unlock it by purchasing credits. Enterprise eligibility requires credit-based or USD usage-based billing; Edu requires a credit-based plan. Legacy rate-limit-based Enterprise agreements are excluded. Enterprise owners must enable access, which is off by default, and existing per-user spending controls continue to apply. These conditions come from OpenAI’s speed documentation.

For a workspace rollout, separate the commercial agreement from the user’s permission. Ask the owner to confirm both before troubleshooting a missing option. A paid account alone does not establish entitlement to every speed tier. Record the workspace, selected model, mode and account route in your rollout notes so another operator can reproduce the configuration.

An initial pilot should use a small group with clear spending ownership. Choose a recurring workflow with a known acceptance test, such as reviewing one saved change against its requirements. Preserve the same inputs and review criteria across modes. This is a suggested evaluation procedure, not a report that we ran a subscription task or measured the service.

Included usage, purchased credits and API dollars

OpenAI’s Work and Codex rate card adds a Sol Ultrafast row: included allowance is consumed at eight times the Standard rate; purchased credits and Enterprise pay-as-you-go use six times. Work and Codex share usage. Pro $500 uses included allowance before available credits. These are billing multipliers, not guaranteed improvements in elapsed task time.

Keep two columns in a budget: allowance consumption and additional-credit spending. Combining them into one multiplier hides which pool a session draws from. A team evaluating a faster mode should also record retries and accepted results, because a shorter unsuccessful run does not represent a useful saving.

The API model reference lists Ultrafast at six times Standard. Short-context rates per million tokens are $12 uncached input, $0.60 cached input, $15 cache writes and $60 output. Above 272,000 input tokens, the corresponding rates are $24, $1.20, $30 and $90 for the whole request. Regional processing adds 10% where applicable. See the official model pricing.

As a text-only arithmetic example, 10,000 uncached input tokens and 2,000 billable output tokens cost $0.24 at short-context Ultrafast rates. The same token counts cost $0.04 at Standard rates. This example excludes tools, caching and regional charges and makes no claim about how many tokens a real task will use. The pricing calculator remains an estimate based on its displayed registry; do not assume it already contains this new mode.

Configure the Responses API and check separate limits

The API Ultrafast guide specifies gpt-6.1-sol with service_tier set to ultrafast on each response.create event. HTTP is supported. OpenAI recommends persistent WebSockets, particularly for agentic applications with frequent tool calls, to reduce connection overhead.

For a first integration check, change the service tier on a saved, bounded request rather than replacing the entire application flow. Keep the prompt, tools, reasoning setting and output budget fixed. Log completion status, token usage and the selected configuration. Then compare the accepted output and observed latency with the Standard run. This proposed comparison has not been executed here.

Ultrafast has separate limits. The guide lists default Sol throughput of 1 million tokens per minute on Build, 4 million on Launch and 40 million on Grow. Check the organization’s actual limits before increasing traffic. A higher published ceiling is not evidence that a particular workload will sustain that throughput, and parallel tool activity still needs its own resource controls.

US and European inference residency

Sol Ultrafast supports inference residency in the United States and Europe, defined for Work and Codex as the EEA plus Switzerland. The API guide also lists US and EU data residency and global processing. Astra Ultrafast remains US-only for inference residency in the product documentation. Match the model and route to the requirement instead of transferring one model’s regional restriction to another.

A residency review should identify the actual endpoint and workspace setting, the approved region and the data handled by the workflow. Record those separately from the subscription name. Before extending a pilot to sensitive documents, confirm the configured route meets the organization’s requirements; a model appearing in a picker does not complete that review.

Decide where faster generation earns its cost

Start with tasks where waiting for generated output is a meaningful part of the workflow. Track time to the first useful output, time to an accepted result and the total charged usage. Those measures answer different questions. Database calls, browser actions and human review can dominate completion time even when generation becomes faster.

Use the new tier selectively after the comparison shows a useful benefit. Keep Standard as a recorded baseline and retain a clear mode label in logs. The release offers another operating choice with explicit access, billing and regional conditions; it does not require changing every task. For the broader distinction between product credits and API billing, use the pricing guide.