May 2026 was the most active month for AI model releases so far this year. Three weeks of launches produced more shipped models than all of Q1 combined. Here is every release that matters, what it actually does, and where you can try it.
The Big Three
Google Gemini 3.5 Flash (May 19)
Released at Google I/O, Gemini 3.5 Flash is now the default Gemini model. It beats Gemini 3.1 Pro on coding benchmarks (76.2% Terminal-Bench) and serves as Google's agentic default. Available in the Gemini app, AI Studio, and through Antigravity.
Why it matters: Google is positioning Flash as the workhorse model — fast, cheap, and good enough for most tasks. If you're building on Gemini, this is your starting point.
Alibaba Qwen 3.7 Max (May 20)
Qwen 3.7 Max ships with a 1M-token context window and scores 92.4 on GPQA Diamond, making it the strongest Chinese model on Artificial Analysis leaderboards. Available through Qwen Chat and Alibaba Cloud Model Studio.
Why it matters: The 1M context window combined with top-tier reasoning scores makes this the model to watch for long-document tasks and codebase analysis. It's also the strongest open-weight competitor to Western frontier models right now.
Subquadratic SubQ 1M-Preview (May 5)
SubQ claims a 12-million-token context window — the largest ever shipped in a production model. Available through API beta and SubQ Code CLI.
Why it matters: 12M tokens is roughly 9 million words. That's enough to load an entire codebase, documentation, and conversation history simultaneously. If the latency numbers hold up, this changes how we think about context limits.
The Rest of the Field
| Model | Date | Key Feature | Where to Try |
|---|---|---|---|
| xAI Grok 4.3 | May 4 | 1M context, native video | grok.com, xAI API |
| Baidu ERNIE 5.1 | May 8 | #4 on LMArena Search Arena | ernie.baidu.com |
| OpenAI Codex Mobile | May 14 | Approve coding tasks from phone | ChatGPT iOS/Android |
| xAI Grok Build 0.1 | May 14 | Agentic coding model (early access) | grok.com |
| ChatGPT Personal Finance | May 15 | Plaid-backed bank integration | ChatGPT Pro (US) |
| xAI Grok Skills | May 18 | Persistent custom expertise system | grok.com |
| Cohere Command A+ | May 20 | 218B sparse MoE, Apache 2.0 | Hugging Face, Cohere API |
| xAI Platform Connectors | May 22 | Vercel, Canva, Gamma integrations | grok.com |
Anthropic Mythos (Still Private)
Anthropic's Claude Mythos Preview was announced April 7 but remains behind Project Glasswing with no public release date. It scored 97.6% on USAMO 2026 — the US Mathematical Olympiad — representing the largest single-generation capability jump Anthropic has ever produced. You can't use it. Neither can we.
What This Means for Model Selection
If you're choosing a model right now:
- Coding tasks: Gemini 3.5 Flash or Qwen 3.7 Max
- Long context: SubQ (12M) or Qwen 3.7 Max (1M)
- Open source: Cohere Command A+ (Apache 2.0) or Qwen 3.7 Max
- Budget: DeepSeek V4 Pro remains the cheapest frontier option
- Reasoning: Mythos leads but is unavailable; GPT-5.5 and Gemini 3.1 Ultra are the best public options
Looking Ahead to June
Google I/O set the pace for the rest of Q2. Expect refinement releases (Gemini 3.5 Pro, potential Mythos wider access) and continued price pressure from DeepSeek and Alibaba. The open-source ecosystem is accelerating faster than the closed-source one — a trend that started in February and hasn't slowed down.
Last updated: May 28, 2026. All benchmark scores and availability are based on officially announced figures.