AI Model Releases of 2026 So Far: The Complete List

Between May 19 and July 16, 2026, the AI labs shipped more than twenty notable models — including four separate flagships with million-token context windows and a 2.8-trillion-parameter open-weight multimodal reasoner. If you blinked in June, you missed an entire generation. This is the complete list of AI model releases in 2026 so far, with real dates, real specs, and a candid take on which ones matter. Bookmark it; we will keep the pattern analysis honest even where the marketing was not.

The 2026 AI Model Releases Timeline

Dates reflect availability in CoreAI's live catalog, which tracks releases as they ship. The table covers mid-May onward — the stretch when 2026's release pace went from brisk to absurd.

July: Kimi K3 and the GPT-5.6 Family Land

July compressed a year's worth of news into two weeks. OpenAI's GPT-5.6 arrived on July 9 as a three-tier family — Sol (premium), Terra (balanced), Luna (efficient), each with a Pro variant and a 1,050,000-token window. The tiering is genuinely useful and genuinely confusing; our Sol vs Terra vs Luna guide untangles it. One day earlier, xAI shipped Grok 4.5 (specs and pricing in our Grok 4.5 guide).

Then July 16 happened: MoonshotAI's Kimi K3, a 2.8-trillion-parameter open-weight multimodal reasoner with a 1,048,576-token context, instantly the most-discussed open model of the year. The same day, Meta shipped Muse Spark 1.1, which quietly accepts text, images, video, audio, and PDFs in one standard-tier model — arguably the sleeper release of the summer.

June: Anthropic's Double Punch and the Open-Weight Surge

Anthropic bracketed June with two 1M-context releases: Claude Fable 5 on June 9 — the premium "mythos-class" model — and Claude Sonnet 5 on June 30, which brought million-token context to the standard tier and promptly became many people's default writing model.

Around them, the open-weight ecosystem surged: Z.ai's GLM 5.2 (June 16) staked out the value tier with a 1,048,576-token window, NVIDIA's Nemotron 3 Ultra 550B (June 4) gave enterprises a serious open heavyweight, Sakana's Fugu Ultra (June 24) joined the premium reasoning club, and MoonshotAI warmed up for K3 with the coding-focused Kimi K2.7 Code (June 12). Public leaderboards like OpenRouter's rankings spent the month reshuffling weekly.

May: The Month Context Windows Went to a Million

Late May set 2026's tone in ten days flat. Google's Gemini 3.5 Flash (May 19) put a 1,048,576-token window in a mainstream fast model. Qwen3.7 Max (May 21) arrived as Alibaba's agent-focused flagship — covered in our Qwen3.7 Max guide — followed by the value-tier Qwen3.7 Plus on June 3, both at 1M tokens. Claude Opus 4.8 landed May 27 with a Fast variant, MiniMax M3 (May 31) made 524K context a budget feature, and xAI's Grok Build 0.1 (May 20) tested the agentic-coding waters. Earlier in the year, the groundwork had been laid by the GPT-5.4 lineup, Gemini 3.1 Pro previews, and Mistral's Small 4 wave — but May is when the arms race went vertical.

What Patterns Do the 2026 Releases Show?

Three trends jump out of the table. First, million-token context is now table stakes: at least eight models on this list ship 1M-class windows, across premium, standard, and even budget tiers. Second, reasoning stopped being a product line — every notable release above is a reasoning model, full stop. Third, agentic coding became its own category, with Kwaipilot's KAT-Coder pair, Grok Build 0.1, Kimi K2.7 Code, and Poolside's Laguna XS 2.1 all targeting autonomous dev workflows rather than chat.

And one meta-trend: release velocity itself. Twenty-plus models in nine weeks means "which model is best" now has a shelf life of about a fortnight — which is precisely why running your own prompts through several models beats trusting any static review.

Which 2026 Models Should You Actually Use?

Cutting through the list: pick Kimi K3 or GPT-5.6 Terra as your daily driver, Claude Sonnet 5 for long documents and writing, GLM 5.2 or MiniMax M3 when cost matters, and Grok 4.5 for quick technical questions. All of them — plus 300 more — are in the CoreAI catalog under one subscription, and the fastest way to form your own opinion is to line two of them up in Compare with a prompt from your actual work. New releases show up in the catalog as they ship, so this list keeps growing without you managing a single account.

Frequently Asked Questions

What are the biggest AI model releases of 2026 so far?

The consensus headliners: MoonshotAI's Kimi K3 (July 16), OpenAI's GPT-5.6 family (July 9), Claude Sonnet 5 (June 30), Claude Opus 4.8 (May 27), and Gemini 3.5 Flash (May 19). Kimi K3 is the standout open-weight release; GPT-5.6 and Sonnet 5 lead the closed flagships.

When was Kimi K3 released and what makes it special?

Kimi K3 shipped on July 16, 2026. It is a 2.8-trillion-parameter open-weight multimodal reasoning model with a 1,048,576-token context window, strong at complex coding, knowledge work, and long agentic workflows.

What is the difference between GPT-5.6 Sol, Terra, and Luna?

They are capability tiers of the same July 9 family: Sol is the premium flagship, Terra the balanced middle, and Luna the efficient tier — all with a 1,050,000-token context and Pro variants. Most people should default to Terra and escalate to Sol for the hardest problems.

Which 2026 release is the best value?

GLM 5.2 (June 16) is the strongest answer: a 1,048,576-token context and solid reasoning at standard-tier cost. MiniMax M3 and Tencent's Hy3 are the picks if you want to go cheaper still.

Where can I try all of these models in one place?

CoreAI carries every model on this list — 300+ in total across all major labs — in one app on iOS, Android, and the web, with side-by-side comparison, file attachments, and a free tier to start.