GPT-5.6 Sol gets a faster lane while OpenAI cuts prices below it

OpenAI's July 30 GPT-5.6 update changes the operating economics around Sol without describing a new Sol model version. API prices for Terra and Luna fell by 20% and 80%, respectively, while Sol's standard price remained unchanged. For Sol, the material change is Fast mode, which replaces Priority Processing and is priced at twice Standard processing. OpenAI says Fast can deliver up to 2.5 times Standard speed with no change in intelligence, making this a service-tier and latency change rather than evidence of a new capability tier.

The timing matters because OpenAI had published an engineering account a day earlier that attributes serving-efficiency gains to GPT-5.6 Sol working within a human-led process. That account says Sol in Codex helped optimize production kernels, contributing to a reported 20% reduction in end-to-end serving cost, and helped run experiments that raised token-generation efficiency by more than 15%. Those are OpenAI's own operational reports, not an independent evaluation, but they provide context for why the company can lower prices for Terra and Luna while offering a premium low-latency path for Sol.

The resulting family split is clearer than a simple cheaper-versus-better comparison. OpenAI positions Sol for the highest-stakes reasoning work, Terra for balanced everyday production tasks, and Luna for high-volume workloads. Its price announcement says Terra costs $2 per million input tokens and $12 per million output tokens, while Luna costs $0.20 and $1.20, and it characterizes Terra as comparable to GPT-5.5 on intelligence benchmarks at half the price. Technical buyers should treat that comparison as a vendor claim and compare task success, latency, token use, and retry rates on their own workloads before moving an agent step from Sol to a lower-cost tier.

AWS documentation shows that the family now has a distinct operational path for teams using Amazon Bedrock. AWS lists Sol as available in two US East regions, while Terra and Luna are also listed for US West, and says all three expose a 272K-token context window, text and image input, text output, the Responses API, and the same set of reasoning-effort levels. Moving an existing OpenAI SDK workload to Bedrock requires changing the base URL, model ID, and authentication approach; it is not merely a billing switch. AWS also notes that classifier-flagged traffic can be retained for up to 30 days for automated abuse detection, a deployment consideration for teams with stricter data-handling requirements.

What is verified here is the published pricing, Fast-mode configuration, and the documented Bedrock integration surface. What remains uncertain is the workload-level tradeoff: a lower token price does not by itself establish a lower cost per completed task, and a faster serving tier does not establish better task quality. A cautious migration is therefore to evaluate the same prompts, tools, reasoning effort, and cache behavior across Sol, Terra, and Luna, then reserve Fast mode for paths where measured latency is worth its premium. The sources support a broader routing strategy, but not a universal replacement rule.