Claude Opus 5, GPT-5.6, Gemini 3.6 Flash: What Actually Changed Since the Last State-of-the-Models Post
Our earlier state-of-the-models rundown is already behind. Between June and early August 2026, all three major Western labs shipped meaningful updates, and the ranking of "which model to reach for" shifted enough to warrant a fresh pass.
Anthropic released Claude Fable 5 on June 9, 2026, then followed it five weeks later with Claude Opus 5 on July 24. Opus 5 isn't a bigger flagship sitting above Fable 5 — it's positioned as a cost-efficient sibling that lands close to Fable 5's frontier intelligence at roughly half the price, with a 1 million token context window and state-of-the-art results on evaluations like Frontier-Bench and GDPval-AA. It's now the default model on Claude Max and the strongest option on Claude Pro, which tells you where Anthropic wants most of its traffic routed.
OpenAI and Google moved on price and efficiency, not just capability
OpenAI's GPT-5.6 went generally available July 9, 2026, with three tiers — Sol, Terra, and Luna. Three weeks later, on July 30, OpenAI cut Luna's price 80% and Terra's 20%, leaving flagship Sol untouched. That's less a capability story than a margin and competitive-positioning one: OpenAI repricing its commodity-facing tier while holding the line on its premium tier.
Google's Gemini 3.6 Flash reached stable general availability on July 21, 2026, priced at $1.50/$7.50 per million tokens with a 1 million token context window, and shipped with roughly 17% fewer output tokens needed on the Artificial Analysis Index compared to Gemini 3.5 Flash — a real efficiency gain rather than a pure price cut. It landed day-one across AI Studio, the Gemini API, the Gemini app, Android Studio, Antigravity, and Vertex AI, signaling Google is treating Flash as the default workhorse tier across its whole developer surface.
Meta took a different route entirely
Meta shipped Muse Spark 1.2 on August 5, 2026, a coding-focused model co-released with Muse Code, a terminal coding agent. Rather than chasing the frontier intelligence ceiling the other three labs are fighting over, Muse Spark 1.2 is priced aggressively at $1.25/$4.25 per million tokens, with an even cheaper contributor tier for developers willing to let Meta train on their prompts and completions — a volume-and-distribution play more than a capability one.
Net effect: the gap between "cheapest usable tier" and "frontier flagship" widened again this summer, and which lab currently holds the efficiency lead versus the capability lead is no longer the same lab two months running.