OpenAI's cheaper GPT-6.1 Sol leaves its new Dots agents on the pricier Astra model

The first person to meet GPT-6.1 Sol is whoever approves the monthly API invoice for a team running coding agents on OpenAI's top model. CNET's DevDay report framed Sol's case that way: business users can run up a huge bill on the latest, most compute-intensive model ^[18].

Sol's rate card has three lines. Fresh input costs $2 per million tokens, cached input $0.10 and output $10 ^[12][13][14]. Output costs five times as much as input, and a cached token costs a twentieth of a fresh one ^[20][21]. A job that resends the same long context on every step is billed mostly at the cheapest line ^[21]. A job that writes long answers is billed mostly at $10 ^[20].

The ratio against Astra comes from OpenAI, and the two reports give different versions of it. CNET has the company claiming Astra-level intelligence at a fifth of the price ^[1]. Mashable has Sol nearly as intelligent as Astra at less than a quarter of the price ^[2]. The implied saving is 80 percent in one account and more than 75 percent in the other ^[16]. Neither report prints Astra's per-token rate, so the ratio cannot be checked against a price list from this material. In my view the word to hold onto is "nearly". A parity claim invites a straight swap of model names in a config file, but a team's users will meet Sol one task at a time. OpenAI said Sol has "exceptionally strong performance on agentic coding, computer use and professional work" ^[15].

Sol's discount stops at the API. Dots are sold through seats ^[3], and seat prices moved against heavy users at the same event. Divided out, the $200 Pro seat now charges $20 for each Plus-equivalent unit of usage, the same as Plus itself and double the old $10 ^[22]. Pro 500 costs two and a half times as much as Pro ^[19].

Speed is sold as a premium too. Ultrafast generates up to 300 tokens per second in Codex and Work, eight times the usual rate, and an API version is coming at six times standard speed ^[6][7]. "Ultrafast is our premium speed tier for workloads where speed matters most," OpenAI said ^[8].

Dots also arrive while the safety argument over agents is still open. CNET sets the launch against concerns over agents "like the kind that hacked HuggingFace, and US and Australian government websites" ^[11]. According to OpenAI, Dots work in the background across a user's context and connected applications, on "a secure cloud computer" ^[4].

Two facts about a team's own bill sort the decision: whether the growth is in tokens or in seats, and whether the work needs Ultrafast. Tokens without a speed need is the case Sol is priced for. The recommendation there is to run Sol against the team's own task set before moving traffic, and the tradeoff is a bet that "nearly" is close enough ^[2]. Tokens with a speed need means paying Astra rates until Ultrafast reaches Sol ^[5]. Seats with light use stay on Plus at $20 a month ^[17]. Seats with heavy use or Dots get nothing from Sol, because Dots run on Astra; the choice there is between cutting usage and paying $500 a month for Pro 500 ^[3][9].