What Is GPT-6.1 Sol? Price, Performance, Who Can Use It
5 min read
OpenAI introduced GPT-6.1 Sol at its Dev Day event on September 29, 2026. The company says the model delivers near GPT-6 Astra intelligence at about a fifth of the cost at standard API rates. Since GPT-6 Sol came out a week earlier, model updates are now measured in weeks, not months.
This article summarizes the GPT-6.1 Sol section of the AI Search channel's weekly AI news video on YouTube and extends it with the price and availability details from the Dev Day announcements and with independent measurements; the video link is at the end. This is a news roundup, and I have not tested the model. Numbers come from OpenAI's announcement, Artificial Analysis measurements and launch-day roundups.
What is GPT-6.1 Sol?
An upgraded version of GPT-6 Sol; according to OpenAI it focuses on agentic coding, computer use and professional tasks. One note: an OpenAI safety executive said the company would not release GPT-6.1 Astra because it did not quite meet the bar of staying within scope and authorization (CBS News). So there is no higher sibling to Sol for now; Astra stays GPT-6 Astra.
| Item | Value |
|---|---|
| Input price | $2 per million tokens |
| Cached input | $0.10 per million tokens (95% discount) |
| Output price | $10 per million tokens |
| Availability | API, Plus, Pro, Business, Enterprise, Edu |
| Artificial Analysis intelligence index | 52 (1 point behind Astra) |
| Cost per task on the index | $0.72 (Astra $3.26) |
Benchmark results
According to OpenAI's announcement and roundups:
- DeepSWE v1.1: Matches Astra's score at roughly a fifth of the cost.
- AutomationBench (medium effort): 2.2 points above Claude Opus 5.5 at about a third of the cost.
- OSWorld 2.0: 2.1 points behind Astra at roughly a seventh of the cost.
- Terminal-Bench Science 0.1: $5.47 per task versus $23.80 for Astra.
- Factual errors: OpenAI reports about 32% fewer factual errors on hard prompts than GPT-6 Sol.
On the independent side, Artificial Analysis puts Sol one point behind Astra and finds its cost per task far lower. By the same evaluator's measurements it gains 12 points on Terminal-Bench 4.0 and 5 on Humanity's Last Exam over GPT-6 Sol, and its hallucination rate drops from 60% to 54%. But it uses 10% to 30% more output tokens than GPT-6 Sol. So the balance of intelligence and cost improves, but there is no free lunch.
The presenter in the video reaches the same conclusion: this is the most cost-efficient frontier model right now (an opinion). Part of the numbers are OpenAI's own measurements; independent checks exist, but the harness used in testing can affect results.
Ultrafast: 300 tokens per second
The other big speed news from Dev Day is the Ultrafast tier: up to 8x faster generation in Codex and 6x in the API, around 300 tokens per second. The price is also 6x: $60 per million input tokens and $300 per million output tokens for Astra. Astra Ultrafast is available on the Pro 500 and Enterprise plans and in the API; GPT-6.1 Sol Ultrafast is "coming soon."
Plan tiers changed
The news that the video summarizes as "the Pro plan's usage is being halved" is this: ChatGPT plans were re-tiered.
| Plan | Usage relative to Plus |
|---|---|
| Plus | 1x |
| Pro 100 | 5x |
| Pro 200 | 10x (previously 20x) |
| Pro 500 (new) | 25x and Ultrafast access |
So the $200 Pro plan's usage was halved, which drew pushback from users. Pro 200 reopened after the September pause in new sign-ups.
Who can use it?
According to the Dev Day announcement, GPT-6.1 Sol is available in the API and on the Plus, Pro, Business, Enterprise and Edu plans. The video says it shows up in the Work tab and in Codex rather than in ChatGPT chat. Check the model picker in your account; access rolls out gradually by plan and region.
Sol, Astra or Claude?
- If cost comes first: 6.1 Sol. The index gap is 1 point.
- If you need peak intelligence: Claude Opus 5.5 or Fable. In the Claude Sonnet 5.5 article I explained Anthropic's prices and effort settings.
- If you need very long output: Gemini 4 Argon outputs 1 million tokens; see What is Gemini 4 Argon.
- If speed comes first: Ultrafast, but at 6x the price.
For tool choice, the vibe coding tools comparison and the same week's OpenAI Dots article may help.
Frequently Asked Questions
What is GPT-6.1 Sol's API price?
$2 per million input tokens and $10 per million output tokens; cached input $0.10.
How far behind GPT-6 Astra is GPT-6.1 Sol?
One point on the Artificial Analysis intelligence index. Its cost per task on the index is $0.72 against Astra's $3.26.
What happened to the Pro plan limit?
Pro 200 dropped from 20x to 10x Plus usage. Pro 100 offers 5x and the new Pro 500 offers 25x.
What is Ultrafast?
A fast-generation tier producing about 300 tokens per second; up to 8x faster in Codex, 6x in the API, and 6x the price.
Is GPT-6.1 Sol free?
No; it is available on paid plans and in the API.
Source
- AI Search, Gemini 4, GPT 6.1, Dots, Claude Sonnet 5.5, Ideogram 4.5, Flux 3: AI NEWS (YouTube, October 4, 2026): the GPT-6.1 Sol section, plan change.
- OpenAI, Introducing GPT-6.1 Sol: the official announcement.
- Latent Space, OpenAI DevDay 2026: Dots, 6.1 Sol, Ultrafast (September 30, 2026): pricing, plan tiers, Artificial Analysis measurements.
- MediaNama, OpenAI launches Dots that keep working without user prompts (October 1, 2026): availability and Ultrafast details.
- CBS News, Sam Altman unveils "dots" (September 29, 2026): the unreleased GPT-6.1 Astra.
Related Posts
Gemini 4 Argon vs Claude Opus 5.5 vs GPT-6 Astra
Gemini 4 Argon leads on legal, finance and long software tasks but trails on terminal and science tasks. Three models compared with a benchmark table.
What Is Claude Sonnet 5.5? Benchmarks, Cost, Free Plan
Claude Sonnet 5.5 is now the free plan's default model. It scores 70.6% on Terminal-Bench 4.0 at $2 / $10 per million tokens. How it compares.
A 3D Game in One Day: Building BOINGWARD with AI Agents
I started from a Unity design pack and switched to Godot the same morning. Parallel Claude Code sessions produced 148 commits and a playable 3D pogo game.