GPT-6.1 Sol vs GPT-6 Astra: Which Model Should Run Your Dot?

Easy to miss amid the dots launch coverage: OpenAI used the same DevDay, September 29, 2026, to announce GPT-6.1 Sol, a separate, cheaper model positioned alongside the flagship GPT-6 Astra — the model that actually drives a dot's judgment. Reported pricing puts Sol at roughly a fifth of Astra's cost per token. That's a meaningful gap, and it's reasonable to ask whether it changes anything about what a dot costs to run. As of this writing, it doesn't — Astra is still the model OpenAI has said powers dots, and nothing in OpenAI's public materials states that Sol has replaced or will replace it there. That distinction needs to stay precise, because the price difference alone makes the opposite assumption tempting.

What Sol is for, if not dots

OpenAI positioned Sol as a lower-cost option for the broader set of ChatGPT and API workloads where Astra's full capability is overkill — the kind of volume use case where a fifth of the price matters more than the marginal capability a flagship model buys you. That's a familiar pattern in the industry: a frontier model for the hardest jobs, and a cheaper sibling for everything else, with the sibling typically trading some capability for dramatically better unit economics. Nothing OpenAI has published characterizes Sol as a downgrade specifically built for dots' use case, and nothing characterizes it as incompatible with it either. The honest state of the public record is: Sol exists, Astra runs dots, and OpenAI hasn't said anything connecting the two beyond having announced them on the same day.

What's actually confirmed about each model's role

Astra's role is the settled half of this comparison: it's the model OpenAI has described as powering a dot's judgment — the reasoning that decides what a dot does next, whether a planned action needs a human sign-off under Custom Rules, and how it interprets the messy, ambiguous instructions a user gives it before walking away. That's a demanding job, and it's reasonable that OpenAI chose its flagship model for it rather than a cost-optimized one. Sol's role is the unsettled half. OpenAI announced it on the same stage, with pricing reported at roughly a fifth of Astra's, aimed at the broader set of ChatGPT and API workloads where a flagship model's extra capability doesn't change the outcome enough to justify the extra cost. Beyond that, OpenAI's public materials don't specify which particular products or query types route to Sol versus Astra today.

Why the price gap is still the interesting part

A dot runs continuously — background monitoring, proactive research, multi-step tasks that can run for hours without a human prompt triggering each step. That's a fundamentally different cost profile than a single chat turn, and it's exactly the kind of workload where a fivefold price difference between models starts to matter at scale, long before it matters to the price OpenAI charges any individual user. Pricing additional dots is already an open question OpenAI hasn't answered. A materially cheaper model sitting right next to the one that currently runs dots is the kind of lever a company holds in reserve exactly when it's trying to answer a pricing question it hasn't finished answering — not evidence that the lever has been pulled.

Astra runs your dot today. Sol exists, costs roughly a fifth as much, and nobody at OpenAI has said what, if anything, that has to do with each other. Both of those sentences are true at once.

What would change this analysis

If OpenAI announces a lower-cost dot tier running on Sol, or confirms that some dot tasks route to Sol while others stay on Astra, that would resolve the open question directly — and would also be the clearest signal yet about how OpenAI plans to price the as-yet-unpriced additional dots and specialist dots it's said will exist. Until then, treating Sol as "the model that might eventually run cheaper dots" is a reasonable hypothesis grounded in the economics, not a confirmed roadmap item. Readers comparing the two models today should do so on the terms OpenAI has actually stated: Astra for dots, Sol for cost-sensitive workloads elsewhere in the ChatGPT and API lineup, with the rest left open.

How this compares to what competitors are doing with model tiers

OpenAI isn't alone in running a two-tier model strategy alongside an agent product. Meta's Muse took the same shape in reverse: it launched in September on the lightweight Muse Spark, built for low latency, before Meta's Superintelligence Lab shipped the heavier Muse Wonder and Muse Wonder Full on October 12 as the capability upgrade. Muse Wonder's own pricing — $4 per million input tokens and $20 per million output tokens through the Meta Model API — sits in a comparable range to what GPT-6 Sol or Claude's Opus 5.5 cost, per Meta's own framing. The difference in direction is notable: Meta started cheap and added a frontier-grade sibling on top, while OpenAI launched dots on its flagship model and added a cheaper sibling alongside it. Neither sequencing tells you definitively where either company's agent product is headed next, but it does suggest both companies see the same two-tier economics as unavoidable once an agent runs continuously rather than per-prompt.

The broader pattern

This is the second changelog-adjacent story this month — alongside Finances in ChatGPT's expansion to Free and Go — where OpenAI has shipped something that plausibly sets up a future dots pricing move without OpenAI confirming the connection itself. Both are worth tracking for the same reason: once the usage grace period on dots ends, OpenAI will need an actual pricing structure, and the pieces for one — a cheap model, a wider financial-data surface — are visibly being assembled in public, one unrelated-sounding changelog entry at a time.

Part 41 of 47 in the ChatGPT Dots series. Previously: Finances in ChatGPT Expands to Free and Go: A Dot Business Model Hint?. Next: Three Weeks In: What OpenAI Still Hasnt Told Dots Users. Browse more builder insights or explore AI skills at aiskill.market.