Daily AI Launch Radar: September 22, 2026

Today’s source-checked launch: OpenAI released GPT-6 Sol for coding and agentic work. The useful question is whether its lower API price justifies a trial on your own workload. OpenAI’s benchmark results are vendor-reported; Kingy has not independently tested this model.

GPT-6 Sol: a lower-cost GPT-6 route

OpenAI made gpt-6-sol available through its API on September 22 and began a gradual rollout in ChatGPT Work and Codex for eligible paid plans. The GPT-6 Sol launch record gathers the source, access and pricing details.

Why it matters: Sol adds another price and capability point for sustained coding and agent tasks. OpenAI positions it below Astra in price. Its published model page lists a 1,050,000-token context window and a 128,000-token maximum output; those limits are capacity, not a measure of answer quality.

Price the request you will actually send

For Standard processing and prompts with at most 272,000 input tokens, OpenAI lists US$2 per million input text tokens and US$10 per million output text tokens. Above that input threshold, its stated multipliers yield US$4 input and US$15 output per million, applied to the full request. Cached input reads and cache writes have their own rates, shown below. Batch, Flex, Fast, regional processing, tool calls and taxes can change the bill.

The figures summarize OpenAI’s published Sol model rates, checked September 22. Recheck the model page before budgeting a production workload.

What to test before choosing it

For an eligible account, compare Sol with the model you already use on the same real task, reasoning effort and quality bar. Track a usable result and the full cost of each run, including long-context and tool charges where applicable. OpenAI’s launch evaluations are useful context, but do not establish that Sol will lead on your tasks. Kingy’s source-based Sol and Luna comparison notes that Sol’s best reported DeepSWE and OSWorld 2.0 scores trail GPT-5.6 Sol’s best reported scores while costing less per task. These are vendor-reported figures, not Kingy hands-on measurements.

Sources and scope

Read OpenAI’s launch announcement for rollout and its model documentation for current specifications and rates. This Radar covers one source-ready launch; it does not imply that other launches were verified or published today. Browse the launch tracker for separately reviewed records.