AI model release news, 2026: the timeline, and the part that does not change
Every roundup on this topic ranks the same four models and stops there. This one gives you the dated 2026 timeline first, then the question those posts skip: when the ordering flips, what do you actually swap? If the answer is your whole tool, you picked the wrong tool.
The 2026 flagships, in release order: Gemini 3.1 Pro and Claude Opus / Sonnet 4.6 (February), GPT-5.4 (March 5), Claude Opus 4.7 (April 16), GPT-5.5 (April 23) and Grok 4.3 (April), Gemini 3.5 Flash and DeepSeek V4-Pro (May 19 and 22), then Claude Opus 4.8 plus the Fable 5 / Mythos 5 pair (June 9). As of this date, Claude Opus 4.8 leads the Artificial Analysis Intelligence Index at 61.4, just ahead of GPT-5.5 at 60.2.
Cross-checked against vendor announcements and artificialanalysis.ai. Numbers move weekly; the structural point below does not.
The 2026 release timeline
Six months, six labs, eleven flagship drops. The cadence is the story: a new top model lands roughly every two to three weeks, and the index leader changed hands at least three times. If you came here asking what shipped in May 2026, it is the Google Gemini 3.5 Flash and DeepSeek V4-Pro rows below.
- Feb 2026Google
Gemini 3.1 Pro
Native multimodal reasoning, 1M token context. Ships inside Google Workspace.
- Feb 2026Anthropic
Claude Opus 4.6 and Sonnet 4.6
The 4.6 pair. Sonnet 4.6 becomes the default fast model for many Claude Code users.
- Feb 2026xAI
Grok 4.20 Beta
Real-time web access and image generation pushed into beta.
- Mar 5, 2026OpenAI
GPT-5.4
Standard, Thinking, and Pro variants under one name.
- Apr 16, 2026Anthropic
Claude Opus 4.7
The frontier Opus step before 4.8. Coding and agentic gains.
- Apr 23, 2026OpenAI
GPT-5.5
Unified system: a standard model, a GPT-5.5 Thinking variant, and a real-time router that picks between them.
- Apr 2026xAI
Grok 4.3
Adds document generation and video input. Best features sit behind the SuperGrok Heavy tier.
- May 19, 2026Google
Gemini 3.5 Flash
Announced at Google I/O 2026 and GA the same day. Google says the Flash tier beats the older 3.1 Pro on nearly all benchmarks. Becomes the default in the Gemini app and AI Mode in Search.
- May 22, 2026DeepSeek
DeepSeek V4-Pro
The launch-window 75 percent discount was made permanent at roughly $0.435 / $0.87 per million input/output tokens, the cheapest frontier-class option of the month.
- Jun 9, 2026Anthropic
Claude Fable 5 and Mythos 5
1M token context, 128k max output, always-on adaptive thinking.
- Jun 2026Anthropic
Claude Opus 4.8
Current frontier Opus. Leads the Artificial Analysis Intelligence Index at 61.4 as of mid-June.
When the leaderboard flips, what do you actually change?
Here is the question the comparison posts never finish. Opus 4.8 is on top today. GPT-5.5 was on top three weeks ago. Something else will be on top in July. So the practical question is not which model wins this benchmark, it is what it costs you to move when the answer changes.
If your coding agent hardcodes a model, the answer is: you wait for the vendor of the tool to ship an update, then you reinstall, then you hope your settings survived. If your tool makes the model a setting, the answer is: you flip a toggle. Those are very different lives, and almost nobody writing about model releases mentions which one you are signing up for.
The model is the news. The harness is what you live in.
Two things change at different speeds. The one in the headlines turns over monthly. The one you spend your day inside should not have to.
| Feature | A model-locked tool | Fazm (harness) |
|---|---|---|
| New flagship model ships | Wait for an app update that adds it | Reachable through your own plan, same day |
| Switching Claude vs Codex | Different app, different setup | Swappable backend per chat |
| Smart vs fast on one task | Usually fixed per session | Opus / Sonnet toggle in the chat header |
| Cost of the newer model | Often a new tier or new bill | Hits your existing Pro or Max plan |
| Routing to a proxy or Copilot | Rarely supported | Custom API endpoint support |
If you only ever use one model and never change it, a model-locked tool is fine. This page is for everyone watching the index move every few weeks.
How Fazm reaches a new model the day it lands
This is the part you can verify, not a marketing line. Fazm is open source, so the mechanism is in the open.
Fazm signs in with your personal Claude account over OAuth (the Bring Your Own Claude flow). The model comes from your plan, not from a key baked into the app bundle, so a new Anthropic model appears through your existing Pro or Max subscription.
The chat header carries a Smart/Fast model toggle, Opus for smart and Sonnet for fast. It is wired through FloatingControlBarState, FloatingControlBarView, and FloatingControlBarWindow. No reinstall to change which model answers.
For anything outside your Claude plan, Fazm takes a custom API endpoint, so you can route through a corporate proxy, GitHub Copilot, or any Anthropic-compatible gateway. The agent loop stays identical; only the destination changes.
Source: the public llms.txt describes the Smart/Fast toggle and the Bring Your Own Claude flow, and the full app is open source at github.com/mediar-ai/fazm.
What does not change when the model does
The model is the part that turns over. Everything below is the part you keep, regardless of whose model is on top in July.
Your sessions
Chats survive a Mac restart. Every window is auto-restored with full history, no matter which model produced it.
Your context
No auto-compacting. The full chat history stays live in context for the lifetime of the window.
One-click forking
Branch any conversation into a new window with the full prior context. The original stays untouched.
Reach beyond the terminal
The same agent drives your real browser, native Mac apps, and Google Workspace via accessibility APIs, not screenshots. Swapping the model does not change what it can touch.
Voice-first input
Hold a hotkey and talk. Same agent loop, no typing, independent of which backend is selected.
Local and open source
Runs on your machine. The mechanism is auditable, which matters for anything that can act inside your logged-in browser.
The honest takeaway
Read the release news. It is genuinely useful to know Opus 4.8 leads today and GPT-5.5 is a hair behind. But do not let a monthly benchmark decide your tooling. The teams that move fastest are not the ones who reinstall every time the index reshuffles. They are the ones who picked a harness where the model is a toggle, kept their sessions and context, and spent the saved time shipping. The news is the model. The decision is the harness.
Want the model to be the only thing you swap?
Walk through how Fazm runs Claude Code and Codex on your own account, with a toggle instead of a reinstall when the next model ships.
Questions people actually ask
Frequently asked questions
What AI models were released in 2026?
In order through mid-2026: Gemini 3.1 Pro and Claude Opus/Sonnet 4.6 in February, GPT-5.4 on March 5, Claude Opus 4.7 on April 16, GPT-5.5 on April 23 and Grok 4.3 in April, then Claude Opus 4.8 and the Fable 5 / Mythos 5 pair on June 9. As of June 20, 2026, Claude Opus 4.8 leads the Artificial Analysis Intelligence Index at 61.4, just ahead of GPT-5.5 at 60.2.
Which 2026 model is best for coding right now?
Claude Opus 4.8 and GPT-5.5 are roughly neck and neck at the top for coding as of June 2026. Gemini 3.1 Pro leads on reasoning and data analysis, and Grok 4.3 is the cheapest of the four. For an agentic coding tool the model choice matters less than how often you can change it, which is the point of this page.
Do I have to reinstall my coding agent when a new model ships?
It depends entirely on whether your tool hardcodes a model or lets you pick one. Fazm runs the real Claude Code agent loop through your own Claude account, with a Smart/Fast toggle in the chat header (Opus for smart, Sonnet for fast). When Anthropic ships a new model to your plan, it is reachable without an app update or a new download.
Can I switch between Claude and Codex without changing apps?
Yes, inside Fazm. It wraps Claude Code via ACP and also bundles Codex (codex-acp) as a swappable backend per chat. You change the backend on a single conversation while leaving your other windows on whatever they were running.
How does Fazm reach a new model the day it launches?
Fazm signs in with your personal Claude account over OAuth (the Bring Your Own Claude flow) and the in-chat model toggle is wired through FloatingControlBarState, FloatingControlBarView, and FloatingControlBarWindow. Because the model comes from your plan rather than a key baked into the app, a freshly released Anthropic model shows up through your existing Pro or Max subscription. For non-Anthropic routing, Fazm also supports a custom API endpoint, so you can point it at a corporate proxy, GitHub Copilot, or any Anthropic-compatible gateway.
Is the leaderboard a good reason to switch tools?
No. The index ordering moves every few weeks (Opus 4.8, then GPT-5.5, then whatever ships next). Switching tools every time the top score changes is expensive. Switching the model inside one tool you already know is cheap. Pick a harness that makes the model the swappable part.
Does Fazm cost extra to use the newest model?
Usage hits your existing Claude Pro or Max plan because you bring your own account. Fazm itself is free to start and fully open source, so a new model release does not come with a new bill from the tool.
Keep reading
Why agent tooling beats model upgrades
The case that the loop you run an LLM in matters more than which LLM is on top this month.
Latest Anthropic API release notes, June 2026
A dated, verified list of what Anthropic shipped to the Claude API in June 2026, annotated for wrapper users.
An AI agent for macOS
What a native Mac agent reaches that a terminal does not: the browser, native apps, and Google Workspace.
Comments (••)
Leave a comment to see what others are saying.Public and anonymous. No signup.