GPT-5.6, Grok 4.5 and rivals cluster within six points

OpenAI shipped GPT-5.6 to general availability on July 9, bundling a flagship called Sol with two smaller siblings, Terra and Luna. Terra is positioned to match GPT-5.5's quality at half the price, and Luna undercuts that further. For anyone paying by the token, that's the number worth tracking, not which model wins a benchmark.

Within days, xAI put out Grok 4.5, Meta dropped Muse Spark 1.1, and Cognition shipped SWE-1.7. The instinct is to ask who won. That's the wrong question.

On the Artificial Analysis index, the top three cluster inside six points: Fable 5 at 59.9, Sol at 58.9, Grok at 54. That's a rounding error wearing a leaderboard costume.

The pattern, not the winner

Decrypt walked through the cluster and landed on the same read AI Explained did. Near-frontier capability arrived at a fraction of prior cost, all at once, from four different labs.

Muse Spark 1.1 is a 1M-context agentic model rivaling GPT-5.5 and Opus 4.8. Sol runs on Cerebras hardware at up to 750 tokens per second.

The exception is Kimi K3, the strongest open-weight release of the month and close to Sol and Fable 5 on quality. Its pricing breaks from the rock-bottom rates defining Chinese open models: cache-hit input runs $0.30 per million tokens, cache-miss $3, output $15.

Simon Willison and Jamin Ball both called it "open weights, closed prices." The floor isn't dropping uniformly anymore.

I've stopped defaulting to the biggest model for routine work in my personal projects. The quality delta doesn't show up in the output anymore. It shows up in the invoice.

My bet: the labs that win from here aren't the ones with the top benchmark score. They're the ones cheap enough that switching away stops being worth the effort.

Kimi K3 pricing near frontier rates instead of undercutting them is the first sign that even open-weight players are chasing margin over adoption. Watch whether the rest of the open-weight field follows.