The model industry had another one of those weeks where the product names sound like a solar system and every chart announces the end of the previous chart. OpenAI has Sol. Google has Argon. The writer still has the email they have been avoiding since breakfast.
OpenAI introduced GPT-6.1 Sol on 29 September 2026. It lists standard API pricing at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens, and positions the model near Astra on several work tasks at lower cost. The launch page places it in the API, ChatGPT Work, and Codex; it specifically says it is not yet in Chat. Launch details and the dated DevDay recap.
Google announced Gemini 4 Argon on 30 September. Its initial rollout is to selected cyber defenders through the Fairwind Program, with broader availability still ahead. The announced introductory API rate is also $2 per million input tokens and $10 per million output tokens; Google lists $4 and $20 after the introductory period. A preview announcement is not general availability. Google's announcement.
Those are provider prices and claims checked on 5 October 2026. They are not TextVeil's prices, an OpenRouter quote, or a declaration that we switched the app to either model.
Cheap generation changes the wrong temptation
For scale, 2,000 standard input tokens and 1,000 output tokens at Sol's published rate cost $0.014 in model tokens. That is arithmetic for one request before tools, retries, and other charges. It says nothing about whether the thousand output tokens deserved to exist.
Cheap generation makes another draft easy to order. It also makes it easy to stop deciding what the draft is for. Ask for five versions, combine the middle of three, ask for polish, then wonder why the result sounds like a panel discussion among assistant managers.
The valuable saving is the version you can use with fewer repairs. If a low-cost model preserves the facts and catches your rhythm, it has done the job. If a flagship model needs three passes to stop inventing emphasis, the expensive badge has bought you more supervision.
The money can move faster than the writing
Menlo advertises a $100 million Anthology Fund built with Anthropic. The capital has a plan for the application layer. That is a fact from Menlo's own site, and it is a useful reminder when another launch tells you the next stage is inevitable. Investors can fund a distribution strategy. They cannot decide which of your sentences is worth keeping.
Our opinion is simple: writers should capture the benefit of cheaper engines without surrendering their style to whichever engine currently has the loudest launch. Keep the brief portable. Keep a clean original. Keep the choices that make the draft recognizably yours.
Give the paragraph something the benchmark cannot supply
A voice is a set of decisions: where you start, what you notice, how long you let a sentence run, and when you stop explaining. Gonzo heat can work. So can a dry, plainspoken line. The mistake is letting every piece drift into the same lubricated middle because that is what the model does when nobody takes the wheel.
That is why TextVeil keeps Humanize and Restyler visible. Humanize gets after the residue; Restyler lets you choose the attitude and movement. Neither needs a fake claim that we defeated every detector on earth. We need your draft to survive the edit, sound less like a server room, and still say what you meant.
The best outcome from a cheaper model week is not more text. It is more room to be particular about the text you keep.