This issue opens as text in your browser. It is not a PDF download. Enter your name and email on Payhip. Just name, not first and last.
The Sol sticker held at $2 / $10. The new cut is the cache read. Past 272K input tokens, the whole request reprices. GPT-6.1 Astra did not ship.
Page 1 ยท The cut is the cache, not the sticker
Last week the Signal was the sticker. This week the sticker held. GPT-6.1 Sol lists at $2 input and $10 output per million tokens, and OpenAI put the new discount on a different line.
The introducing page leads with "Near-Astra intelligence for a fifth of the price." Under that: GPT-6.1 Sol "nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard input and output token prices. Cached input costs just $0.10 per million tokens, 95% less than standard input pricing and 50% less than GPT-6 Sol's cached input pricing." [VERIFIED, openai.com/index/introducing-gpt-6-1-sol]
Illustration ยท reused context as a closed loop of cyan light. Not a chart.
The model card matches the sticker and adds the write price. Per 1M tokens, standard, at or under 272K input: input $2.00, cached input $0.10, cache writes $2.50, output $10.00. Astra's card: input $10, cached input $1, cache writes $12.50, output $50. One-fifth of $10 and $50 is $2 and $10. The cache-read gap is wider than one-fifth, $0.10 against $1. [VERIFIED, developers.openai.com model cards for gpt-6.1-sol and gpt-6-astra]
OpenAI's own evals, not an independent bench: on DeepSWE v1.1 they say Sol matches Astra at roughly one-fifth of the cost and beats GPT-6 Sol by 6.4 percentage points. On Terminal-Bench Science 0.1 at maximum effort they report $5.47 average cost per task for GPT-6.1 Sol against $23.80 for Astra, with Astra still the highest score they list, 68.1%. Company figures. Your invoice is [UNMEASURED] until you run the same work. [VERIFIED as OpenAI's published claims]
Thesis: this is not another headline cut. The Sol input and output rates did not move. The discount they printed against their own previous Sol is the cache read, which they say is 50% lower. If the agent does not reuse context, it is not in the sale.
Page 2 ยท 272K reprices the whole request
The card states the cliff in one sentence: "Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request." The whole request, not a surcharge on the overflow. Applied to the standard rates, that is input $4, cached input $0.20, cache writes $5, output $15 per 1M tokens. [VERIFIED multipliers. The dollar line is those multipliers applied to the verified standard rates, not a second quoted table.]
Illustration ยท the long-context schedule as a bridge that ends.
Fast mode is 2x standard. Batch and Flex are 50% lower than standard. The introducing page says GPT-6.1 Sol is in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu, in the API as gpt-6.1-sol, and "not yet available in Chat." [VERIFIED]
Thesis: a long agent that crosses 272K input tokens does not pay the fifth-price sticker. It pays the long-context schedule on every token in that request. The cache discount and the cliff sit on the same card.
Page 3 ยท The model they benched, and one move
The other GPT-6.1 did not become a product. BBC reports OpenAI will not release GPT-6.1 Astra over safety concerns. Saachi Jain, OpenAI's head of safety systems, said it "didn't quite meet the bar," on "staying within scope and authorisation and how it communicates back to the user about the type of work it's done." CNBC, 28 September 2026, confirmed the non-release and quoted Jain: it "didn't quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it's done." [VERIFIED as BBC and CNBC quoting OpenAI's safety lead. A first-party OpenAI post announcing the shelf: UNAVAILABLE this pass.]
Illustration ยท a model left on the bench. No portrait, no mark.
Do not budget a model that was held for scope. The one you can call is the one with the cache line and the 272K cliff.
One move in 24 hours:
- On one production agent, split yesterday's tokens into fresh input, cached input, and cache writes. If you cannot see cache hits, the fifth-price claim on your stack is UNMEASURED.
- Count requests whose input crossed 272K. Those are full-request reprices, not sticker math.
- Leave GPT-6.1 Astra out of the budget. It is not a SKU this week.
The Signal is which line on the card you actually pay, not the fifth-price headline.
Next paid step stays the Field Kit ($17). Launch price: $12 with code LAUNCH5 through Oct 25 (first 20). This issue stays free and opens as text, not a PDF: payhip.com/b/4tHmb.
News feed dzn-news-feed--dylanad2045.replit.app was not live (UNAVAILABLE). dyl-nawful posts.json 404 (UNAVAILABLE). Today's @dylan2045ad posts were not re-quoted, X API spend cap (UNAVAILABLE). No metrics from those posts are used.
