OpenAI’s new GPT-6 models undercut Claude and DeepSeek on price

OpenAI just cut prices again, and this time it’s aiming straight at Anthropic and DeepSeek’s wallets.

The new GPT-6 Sol and GPT-6 Luna models are both 50 percent cheaper than their previous versions. Sol now costs less than Claude Opus 5.5 to run, whilst Luna undercuts even DeepSeek’s V4.1 Flash, long seen as the budget king. We’re talking $2 per million input tokens for Sol and just $0.10 for Luna. That’s a proper price war, not a token gesture.

The catch: cheaper doesn’t mean better

Here it gets messy. Claude Opus 5.5 still tops the Artificial Analysis Intelligence Index with a score of 58, well ahead of Sol’s 48. DeepSeek’s V4.1 Flash also still beats Luna on coding benchmarks. So OpenAI’s pitch isn’t “best model,” it’s “good enough, dirt cheap.”

The models are safer too. Sol’s hallucination rate dropped from 92 to 60 percent, and Luna’s from 93 to 77 percent, according to the AA-Omniscience Index. Sol pulls this off mostly by refusing to answer more often, which also drags its accuracy down a touch. Luna’s accuracy barely moved. Both models also slipped on GDPval-AA v2.1, often turning in shorter, incomplete work. Sol even fell behind its predecessor on coding and debugging tasks.

None of this is stopping the pace. Anthropic dropped Claude Opus 5.5 today too, and OpenAI is clearly determined not to let price be the reason anyone picks a rival. Whether raw intelligence or a bargain API key wins this round is still up for grabs, but for now, OpenAI’s betting on the latter.

For more updates like this, check out the gaming news section!