The vendor page at openai.com/api lists GPT-6 Astra at $10.00 per 1M input tokens and $50.00 per 1M output tokens. The same page lists GPT-6.1 Sol at $2.00 and $10.00, and GPT-6 Luna at $0.10 and $0.50. All three tiers show 1.05M context length and 128K max output tokens. The cut-offs are Apr 30, 2026 for Astra and Sol, and May 18, 2026 for Luna.

The Decoder desk added a human-reaction frame and a mass-release story, but the API page does not list manuscripts, proof counts, or a release date for mathematical work. It names GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna as priced API models, not as a proof repository. The pricing note says standard rates apply to context lengths under 272K, a limit the wire did not carry. That omission matters because the vendor page is a pricing sheet, not a mathematical archive.

Die Brief reads the page as a cost ladder, not a capability claim. The buyer is constrained by token price, because the three tiers share the same context and output ceiling while their input and output rates differ. The document does not prove that Astra, Sol, or Luna can solve open mathematics, verify proofs, or replace a human reviewer. For the fab or the cloud operator, the listing says nothing about silicon, memory, or inference cost per token.

What remains unmeasured is the quality of any mathematical output, the cost above 272K context, and the latency or throughput of the API. The sheet gives no benchmark, no proof audit, and no hardware or training detail. A buyer can price a call, but cannot price the value of a verified result. The 128K output figure is not defined as a hard cap or a default.

The API page turns the launch into a price ladder, so the buyer's constraint is token cost, not context. It does not prove mathematical capability, proof validity, or inference cost.

The document leaves out benchmark results, proof verification, latency, throughput, hardware, and pricing above 272K context.

The vendor sheet fixes three price points at identical 1.05M context and 128K output, so cost per token is the only differentiator. It leaves capability, proof verification, and >272K pricing unmeasured.