80% Cheaper. OpenAI's New Model Still Trails Astra.

Three model launches in eight days. Two of them OpenAI's own.

On September 29, OpenAI rolled out GPT-6.1 Sol, a week after launching GPT-6 Sol. That first launch wasn't solo either: Anthropic released Claude Opus 5.5 around the same time, then OpenAI shipped GPT-6 Sol and GPT-6 Luna. This is not the cadence of a lab that owns an uncontested lead.

Here's the actual pitch: "We're introducing GPT‑6.1 Sol, an upgrade to GPT‑6 Sol that 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."

Translation: we didn't build something smarter than our own flagship. We built something almost as good, for a fifth of the money, because the thing under pressure wasn't the capability ceiling. It was the invoice.

The Price Cut Wearing a Capability Costume

The numbers confirm it. GPT-6 Astra costs $10 input, $50 output; GPT-6.1 Sol costs $2 input, $10 output; GPT-6 Luna costs $0.10 input, $0.50 output. The real headline is cached input: $0.10 per million tokens, 95% less than standard pricing.

On the signature coding benchmark, GPT-6.1 Sol matches Astra at roughly one-fifth the cost, while beating GPT-6 Sol's score by 6.4 percentage points at lower reasoning effort. Per-task economics: $5.47 for Sol versus $23.80 for Astra and $23.21 for Claude Opus 5.5, per OpenAI's evaluation.

Notice what's missing: a claim that Sol beats Astra at anything.

On the hardest work, Astra still wins. Terminal-Bench Science: Astra at 68.1%. Every number above is OpenAI grading its own homework. One independent writeup put it plainly: the benchmark claims are unpublished outside OpenAI and have not been reproduced by anyone else.

That's not a scandal. It's the asterisk the "new frontier model" headlines keep dropping.

Who Pays, Who Profits

GPT-6.1 Sol shipped day-one into Microsoft Foundry and AWS Bedrock. That distribution matters more than the benchmark deltas. It lands directly on enterprise accounts already budgeting per-token costs for coding agents—the exact customers who'd otherwise price out Claude or an open-weight model for the same job.

Instead of asking customers to pay more for more intelligence, OpenAI is asking them to keep paying OpenAI for slightly-less-than-Astra intelligence at a price that makes switching look like more work than it's worth. The September price war context backs this up: GPT-6 Sol and Luna already launched at half the price of their predecessor tier before this release landed.

The consensus will read "frontier model" and conclude OpenAI's lead is intact.

The honest read: OpenAI just told you, in its own pricing table, that the lead is expensive to maintain and getting cheaper to challenge.

What to watch: whether Anthropic or Google ship a comparable price cut inside 60 days. Whether independent benchmarks—not OpenAI's—confirm the Astra-parity claims. Whether any enterprise account visibly moves workloads to a competitor now that Sol removed the "it's too expensive to switch" excuse.