OpenAI, Meta, SpaceXAI Undercut Anthropic on Model Pricing

Three frontier model releases in a single week, and every one of them opened on price rather than capability. Bloomberg reports that OpenAI's top-tier GPT-5.6 Sol scored 88.8% on the Terminal-Bench 2.1 coding benchmark, edging Anthropic's Fable5 at 88%, and landed within one point on the Artificial Analysis Index at 59 versus 60. The number worth staring at is the cost to run the full test set: $1,087 for Sol against $2,315 for Fable5. Half the bill for effectively the same score.

Meta released Muse Spark 1.1 the same day, priced at $1.25 per million input tokens and $4.25 per million output tokens, roughly one-tenth of Fable5's $10 and $50. SpaceXAI's Grok 4.5, released a day earlier, sits at $2 and $6. GPT-5.6 itself ships as three tiers, Sol, Terra, and Luna, which OpenAI says is designed to complete more work while using significantly fewer tokens.

The reason this matters for anyone signing an API contract right now: Anthropic has been widely viewed as the frontrunner of the moment, and Fable5's price reflected that. Three rivals in one week have effectively said the premium is not defensible at the same capability tier. If your finance team is asking why the AI line item keeps climbing, you now have three concrete alternatives to point at with published benchmarks and published prices.

The honest caveat is that a benchmark win is not the same as being reliable on a real workload, and Fable5 still leads Sol on the Artificial Analysis Index the labs actually cite. What the reporting does not give you is Anthropic's response, whether Fable5 pricing gets cut, whether a cheaper tier ships, or whether the strategy is to sit on capability leadership and let the price gap stand.

For enterprise buyers, this is the strongest negotiating position they have had in this cycle. For the labs racing to catch Anthropic, cost efficiency is the wedge that has finally got customers listening.