Flagship Model Release Roundup: Summer 2026
OpenAI, Anthropic, and Google each shipped a major model within five weeks of each other this summer — here's what each one is actually for.
Summer 2026 was dense for frontier AI releases. OpenAI, Anthropic, and Google each shipped a significant new model within about five weeks of one another, and each one is aimed at a noticeably different job rather than competing head-on for the same use case.
OpenAI led with GPT-5.6 on July 9, releasing it as a three-tier family: Sol for the hardest problems like complex coding and security research, Terra for high-volume business tasks such as support and document analysis, and Luna for fast, cheap, everyday work like summarization and drafting. All three share a 1-million-token context window, and OpenAI cut Sol's API pricing by over 20% in late August after initially limiting access to a small set of preview partners.
Anthropic followed on July 24 with Claude Opus 5, positioned as a daily-driver rather than a narrow specialist — Anthropic's own announcement describes it as coming close to the capability of its own top-tier Fable 5 model at half the price, and it's now the default model on Claude Max and the strongest available on Claude Pro. It leads on software-engineering and agentic-reasoning benchmarks specifically, at $5 per million input tokens and $25 per million output tokens.
Google's release, Gemini 3.7 Flash on August 13, is a different kind of announcement: it's not Google's true flagship — the Gemini Pro line remains delayed with no announced date — but its newest workhorse model, built for coding and agent workflows, with introductory pricing through the end of 2026 before it roughly doubles.
This roundup leaves out entrants from Chinese labs, including DeepSeek's hosted products, that otherwise shipped notable releases over the same period — not for a technical reason, but because this site doesn't recommend hosted AI services controlled by or affiliated with the Chinese Communist Party. That's a standing site policy, not a verdict on any single release; a locally run, offline instance of an open-weights model from any lab is a separate question this policy doesn't cover.
GPT-5.6 Sol
OpenAI — released July 9, 2026
- State-of-the-art on complex coding and security-research benchmarks
- 1-million-token context window
- API pricing cut over 20% as of late August
- Launched to a small set of preview partners before wider availability
- Priced and tuned for heavy, hard problems rather than casual daily use
Claude Opus 5
Anthropic — released July 24, 2026
- Near-frontier performance at roughly half the price of Anthropic's own top-tier model
- Leads on software-engineering and agentic benchmarks
- Now the default model on Claude Max
- Not Anthropic's outright most capable model when Fable 5 is available
- Fast mode doubles the price for lower latency
Gemini 3.7 Flash
Google — released August 13, 2026
- Meaningful accuracy jump on coding benchmarks over the prior Flash release
- Introductory pricing locked in through the end of 2026
- Available across three separate Google developer and enterprise surfaces
- Not Google's true flagship — the Gemini Pro line remains delayed
- Pricing roughly doubles once the introductory period ends
Source: OpenAI — GPT-5.6: Frontier Intelligence That Scales With Your Ambition