All posts

· By Denis Suciu

GPT-6.1 Sol is now in PocketPhantom

OpenAI's newest model is now in PocketPhantom's model picker, on every paid plan. What GPT-6.1 Sol is good for, how to pick it and why Free does not have it.

The PocketPhantom logo above the words “GPT-6.1 Sol” and “Now in PocketPhantom”, on dark glitch art.

OpenAI released GPT-6.1 Sol, its newest model, in late September 2026. It is now in PocketPhantom's model picker, next to Claude, the other GPT models and GLM. There is nothing to set up and no API key of your own to bring.

What it is good for

Reasoning, code, research with the web and long documents. OpenAI says it comes close to GPT-6 Astra, OpenAI's strongest model. It costs a fifth of Astra's price, so it uses much less of your monthly AI.

Like every GPT model in PocketPhantom, it reads text, images and PDFs, up to four files in a message. Web search is on unless you switch it off, and an answer names the pages it used.

What OpenAI measured

OpenAI reports these results from its own tests, against GPT-6 Sol, the model before it, and GPT-6 Astra:

  • DeepSWE v1.1, software work in real codebases: 6.4 points above GPT-6 Sol's best score, and level with GPT-6 Astra at about a fifth of the cost.
  • GDP.pdf, professional work with complex PDFs: close to Astra, at about a fifth of Astra's cost.
  • AutomationBench, workflows in sales, marketing, finance and HR: 4.8 points above GPT-6 Sol.
  • Terminal-Bench Science 0.1, science research tasks: more than twice GPT-6 Sol's score. Astra keeps the top score, 68.1%.
Bar chart of the average cost per task on Terminal-Bench Science 0.1 at maximum effort: GPT-6.1 Sol $5.47, Claude Opus 5.5 $23.21, GPT-6 Astra $23.80.
Average cost per task on Terminal-Bench Science 0.1, each model at maximum effort. Numbers from OpenAI.
Bar chart of the share of answers with a factual error on hard questions at “xhigh” effort: GPT-6 Sol 4.5%, GPT-6.1 Sol 4.1%, GPT-6 Astra 4.0%.
Share of answers with a factual error on hard questions, at OpenAI's “xhigh” effort. Lower is better. Numbers from OpenAI.
Bar chart of API prices per million tokens. GPT-6.1 Sol: input $2, cached input $0.10, output $10. GPT-6 Astra: input $10, cached input $1, output $50.
Standard API price per million tokens. GPT-6.1 Sol costs a fifth of Astra, and a tenth for cached input. Prices from OpenAI.

OpenAI ran some of these tests at maximum effort or at “xhigh”, and some with the model working in a terminal or in a whole codebase. PocketPhantom sends GPT-6.1 Sol at OpenAI's “high” at most and gives it no terminal or codebase to work in, so results in PocketPhantom can differ.

Sources: OpenAI's page for GPT-6.1 Sol and npaka's summary of the announcement, in Japanese.

Which plans include it

Every paid plan includes it: Pro at $20 a month, Power at $60 and Max at $100 or $200. The plans differ in how much AI use they include each month. A bigger model or a higher effort uses more of it, and no plan is unlimited. Compare the plans.

The Free plan keeps to the lighter models, so it does not include GPT-6.1 Sol. On Free it shows a lock and the words “Comes with a paid plan”, and choosing it opens the plans. If you had picked it before, GPT-6 Luna answers in its place.

How to pick it

  • Open the model pill in the chat bar: the button with the model's mark. On a phone it is the Model row in the plus menu.
  • Type 6.1 to narrow the list and choose GPT-6.1 Sol under Featured. The arrow keys and Enter work too.
  • This browser remembers your choice.

Effort

Effort is how long the model may think before it answers. The sliders button in the chat bar opens the effort rail; on a phone it is the Effort row in the plus menu. Pick Low, Medium, High or Max. On GPT-6.1 Sol, Max thinks as long as High. Medium is the default and right for most things. Longer thinking uses more of your monthly AI.

Models and effort in the docs describes every model and every level.

GPT-5.4 mini and nano have left the picker

GPT-5.4 mini and GPT-5.4 nano are no longer in the picker. If one of them was your choice in this browser or in a scheduled task, GPT-6 Luna now answers in its place: OpenAI's small, cheap GPT-6, on every plan, Free too.