---
title: "GPT-6.1 Sol Pricing: Standard, Fast and Ultrafast Tiers Compared"
description: "GPT-6.1 Sol costs $2 input and $10 output per million tokens. Fast is 2x, Ultrafast 6x. Compare tier prices, Codex multipliers and when each is worth it."
url: https://cdnping.org/en/blog/gpt-6-1-sol-codex-default
language: en
published: 2026-10-19
category: "Model watch"
tags: ["GPT-6.1 Sol","Codex","OpenAI API","Ultrafast","multi-agent"]
source: cdnping.org
license: Free to quote and summarize with a link to the url above.
---
# GPT-6.1 Sol Pricing: Standard, Fast and Ultrafast Tiers Compared

On September 29, OpenAI released GPT-6.1 Sol (API model name `gpt-6.1-sol`). It is positioned for complex coding and professional tasks at a lower price than GPT-6 Astra; the Codex changelog describes it as "close to Astra on complex work, at lower cost than Astra." On October 8, OpenAI added Ultrafast mode, which trades a higher price for faster output. Below, the prices, multipliers and use cases are laid out from the official docs. Information is current as of October 2026; the official pages are authoritative.

## What else happened on September 29?

According to the [OpenAI API changelog](https://developers.openai.com/api/docs/changelog), three things shipped that day:

- GPT-6.1 Sol went live, with tool calling through the Responses API.
- Multi-agent delegation (beta): within a single Responses API request, the model can hand work off to subagents.
- The Agents API gained computer use: an agent can complete tasks in an OpenAI-hosted browser, while site access authorization and sign-in are handled by your application.

On the Codex side, the [Codex changelog](https://learn.chatgpt.com/docs/changelog) entry for CLI 0.159.1 states that GPT-6.1 Sol was added as the default model in the built-in model catalog. When GPT-5.5 was retired in Codex on October 14, OpenAI also pointed users to `gpt-6.1-sol`. For migration details, see [GPT-5.5 retired in Codex: how to migrate scripts and scheduled jobs](/en/blog/gpt-5-5-retired-in-codex-migration) and [Codex retires its old default model on October 14](/en/blog/codex-model-retirement-cli-0-163).

## How much does the GPT-6.1 Sol API cost?

From the [official pricing page](https://developers.openai.com/api/docs/pricing), per million tokens, for inputs up to 272K:

| Tier | Input | Cached input | Cache write | Output |
|---|---|---|---|---|
| Standard | $2.00 | $0.10 | $2.50 | $10.00 |
| Batch / Flex | $1.00 | $0.05 | $1.25 | $5.00 |
| Fast | $4.00 | $0.20 | $5.00 | $20.00 |
| Ultrafast | $12.00 | $0.60 | $15.00 | $60.00 |

A few notes on how to read this. Fast is 2x Standard (the pricing page notes that Priority was renamed Fast on July 30), Ultrafast is 6x, and Batch / Flex is 0.5x. Above 272K input tokens, long-context pricing applies: Standard becomes $4 input, $0.20 cached input, $5 cache write and $15 output, and the other tiers rise the same way; see the pricing page for specifics. Regional processing (data residency) and FedRAMP endpoints add another 10%; check the pricing page to see whether that applies to you. For the price structure of the sibling models, see [GPT-6 Sol / Luna on the API: price table and how cache billing works](/en/blog/gpt-6-sol-luna-api-price).

## How do you turn on Ultrafast?

Since October 8, you set `service_tier` to `"ultrafast"` for `gpt-6.1-sol` in the Responses API. It shortens the interval between generated output tokens. OpenAI says it is available to all API customers (subject to rate limits) and supports global processing as well as US and EU data residency.

```json
{
  "model": "gpt-6.1-sol",
  "service_tier": "ultrafast",
  "input": "..."
}
```

Note that it speeds up generation, not the initial connection. Network-level latency (connection setup, round trip, first token) should be measured separately; the method is in [Breaking down AI API latency: connection setup, gateway round trip and time to first token (TTFT)](/en/blog/api-latency-connect-rtt-ttft).

## What are the quota multipliers in Codex?

The multipliers from the [Codex pricing page](https://learn.chatgpt.com/docs/pricing):

| Mode | Quota use within a subscription | Purchased credits / Enterprise pay-as-you-go |
|---|---|---|
| Fast | 2.5x | 2x |
| Ultrafast | 8x | 6x |

The Standard credit rate per million tokens is 50 for input, 2.5 for cached input and 250 for output. On Plus and standard Business plans, local messages run roughly 15 to 160 per five hours; OpenAI says this is an estimate, not a fixed cap. Ultrafast consumes subscription quota at 8x, so the same quota covers noticeably fewer tasks. Check the usage display in your own account for actual numbers.

In Codex, Ultrafast is available only on the Pro $500 tier and eligible Enterprise / Edu plans. It is off by default for Enterprise and must be turned on by the workspace owner (see the October 8 entry in the [Codex changelog](https://learn.chatgpt.com/docs/changelog)). For the full breakdown of subscription tiers and quotas, see [ChatGPT Pro splits into $100/$200/$500 tiers](/en/blog/chatgpt-pro-500-plan-codex-limits).

## When is Fast or Ultrafast worth it?

First, a worked example with a hypothetical task: one request with 1 million uncached input tokens and 100K output tokens (an illustration, not a measurement).

- Standard: $2 + $1 = $3
- Fast: $4 + $2 = $6
- Ultrafast: $12 + $6 = $18

You can decide with this checklist:

1. Is someone waiting on the result? For interactive debugging and edit-as-you-go work, Fast is worth it. For overnight batch jobs, use Batch / Flex at half the price.
2. Is the bottleneck generation or reading context? Ultrafast shortens the gap between output tokens. It does not help with waits caused by long inputs or long tool chains.
3. Is your cache hit rate high? Cached input also scales with the multiplier, though from a small base.
4. Run on Standard for a week first, note average duration and spend, then decide which kinds of tasks to upgrade. Don't enable it globally by default.
5. Make `service_tier` a configurable setting in your scripts so you can switch per task.

## How do you watch latency after launch?

In the first few weeks after a new model ships, availability and latency often fluctuate. cdnping's [/en/status/openai](/en/status/openai) probes the OpenAI API entry point from Cloudflare's edge every 5 minutes, [/en/models](/en/models) records availability and latency for providers on OpenRouter, and you can subscribe to [/en/changes](/en/changes) for price and listing changes. For tracking methods, see also [How to find out right away when an LLM API price changes](/en/blog/track-llm-price-and-model-changes).

## Sources

- [OpenAI API changelog](https://developers.openai.com/api/docs/changelog)
- [OpenAI API pricing](https://developers.openai.com/api/docs/pricing)
- [Codex / ChatGPT changelog](https://learn.chatgpt.com/docs/changelog)
- [Codex pricing and usage](https://learn.chatgpt.com/docs/pricing)