The July Model Wave's Real Signal Is a Price Collapse: Terra at Half GPT-5.5's Cost, Luna Cheaper Still, Grok 4.5 and Muse Spark Near-Frontier

Decrypt·medium signal

Coverage of the GPT-5.6 / Grok 4.5 / Muse Spark 1.1 / Cognition SWE-1.7 cluster converges on one point: within ~48 hours four labs delivered near-frontier performance at a fraction of previous cost, with OpenAI's Terra positioned to match GPT-5.5 at half the price and Luna cheaper still. For builders, the practical consequence is that 'good enough frontier' inference is getting commoditized, shifting the differentiator to orchestration, latency and context length rather than raw model quality. Expect margin pressure on anyone reselling model access.