Between September 1 and October 13, every major lab shipped something: Fable 5.1, GPT-6 Astra, Muse Spark 1.3, Gemini 3.8 Flash, DeepSeek V4.1-Flash, Grok 4.7, Opus and Sonnet 5.5, GPT-6.1 Sol, Mistral Large 4 preview, Haiku 5.5, Muse Wonder preview, and Gemini 4 Argon GA. If your contracts renew this quarter, here is the short version.

Migrate now

Claude Opus 5.5 ($4/$20, 1M context) is the default for agentic coding. Grok 4.7 ($2/$6, 500K context) is the value frontier pick at last generation's price. DeepSeek V4.1-Flash (MIT weights, ~$0.15/$0.60) owns the self-hosting slot. Gemini 4 Argon is GA and safe, with one caveat below.

Wait, deliberately

Three pending events argue for short contracts rather than annual locks: Mistral Large 4's weights (promised within days), the Gemini 4 Pro tier (expected late Q4), and Muse Wonder's GA with unpublished pricing (Q4–Q1). Meta's Muse Spark 1.3 is good today, but Wonder reframes Meta's whole line when it lands.

And one non-event to stop waiting for: Llama 4 Behemoth is indefinitely delayed after Meta's Muse pivot. Our Model Guide marks the Llama line Legacy.

Mind the pricing calendar

Three dates belong on your finance team's radar: Gemini 4 Argon steps up from $2/$10 to $4/$20 on December 1, Gemini 3.8 Flash doubles from $0.75/$3.75 to $1.50/$7.50 on January 1, 2027, and GPT-5.6 Sol's promo is guaranteed only through at least November 21. All three are covered in our pricing alerts. The full badge-by-badge reasoning is in the Model Guide.