AI firms cut token costs as buyers push for cheaper models
What's new: OpenAI, Meta and SpaceXAI rolled out models built to do more work per token as companies rein in AI bills.
The AI industry is shifting from a race for ever-larger models to a fight over cost and efficiency as corporate customers scrutinize spending. OpenAI, Meta and SpaceXAI each released or promoted models in the past week with a key selling point: lower token use and cheaper operation. OpenAI said GPT-5.6 can complete more work with significantly fewer tokens, while SpaceXAI billed Grok 4.5 as twice as token-efficient as comparable rivals. Meta is also pushing aggressive pricing for Muse Spark 1.1. The change comes after businesses embraced heavy AI use earlier in 2026, then tightened limits as usage-based pricing produced unexpectedly large bills.