OpenAI's GPT-6 Sol and Luna cut prices in half, intelligence flat

This launch follows OpenAI's pattern of iterative model releases rather than a leap in capability. The company attributes the price cuts to improvements in caching and inference, saying it is passing those savings directly to users. That framing puts GPT-6 in direct competition with cheaper open-weight models, and OpenAI's benchmark selection — which includes comparisons to Anthropic's Claude lineup but omits metrics like GDPval and Terminal-Bench 4.0 — suggests the company is emphasizing cost-to-performance over raw intelligence gains.

The stakes here hinge on whether the price advantage translates into real-world usefulness. Artificial Analysis found that intelligence scores stayed flat, with regressions on GDPval-AA v2.1, a benchmark for computer-based knowledge work across 44 professional fields. Those regressions were traced mainly to lower presentation quality and incomplete results, which could matter to teams relying on the models for document generation or data analysis. At the same time, the distinction between OpenAI's own models is becoming harder to parse: Luna at maximum effort matches Sol at "xhigh" while costing 78 percent less, a comparison that may confuse buyers trying to choose within the same product line.

The missing response to Anthropic's Opus 5.5 launch is also notable. OpenAI appears to have missed that release, which is potentially up to 40 percent cheaper than Opus 5 with significantly better performance. Whether GPT-6 Sol and Luna can hold their ground may depend on how quickly OpenAI updates its benchmarks and whether enterprise customers prioritize cost savings over the intelligence gains that rivals are still delivering.