Meituan, the Chinese services company, released LongCat-2.5, a free-to-download AI assistant that can read images and handle long coding and computer tasks.

Hugging Face ↗Hugging Face ↗

Not confirmed yet

Spotted early

All early signals

Early signal: A hint that a model is coming before it's announced: a name spotted in a lab's systems, an anonymous test model, or code that adds support for it. Usually right, sometimes not.: hints that a model is on its way before its maker announces it. Usually right, sometimes not, so they're kept apart from the releases until they're confirmed.

Inkling 2 Small2h ago
Other lab · seen in OpenRouter

The last 24 hours

News

01
Anthropic releases Claude Fable 5.2 and Mythos 5.2, priced 20% below 5.1
Anthropic launched Claude Fable 5.2 for broadly available use and Mythos 5.2 for verified organizations, at $8 per million input and $40 per million output Tokens: The small chunks of text a model reads and writes, about three-quarters of a word each. Prices and limits are counted in tokens.. Anthropic says the models lead on coding, Reasoning model: A model that thinks step by step before answering. Slower, but much better at maths, code and tricky questions. and vision, and Artificial Analysis scores Fable 5.2 at 65, its highest so far.
6h ago
02
Nvidia CEO Jensen Huang admits affair with suspected Chinese spy; board opens probe
Nvidia's board opened an investigation into Jensen Huang's relationship with Alice Huang, whom US intelligence reportedly classes as a suspected Chinese informant. Nvidia shares fell about 4-5% and Trump said the FBI is looking into the matter while saying Huang should stay on as CEO.
23h ago
03
Report says Commerce probe of Nvidia-linked Megaspeed chip smuggling moved to DOJ after Nvidia intervened
According to the New York Times, a Commerce Department investigation into Megaspeed for allegedly smuggling Nvidia chips to China was transferred to the DOJ after Nvidia intervened. The Megaspeed story is tied to the Huang scandal, and Bloomberg reports Megaspeed's Nvidia imports top $4.6B.
12h ago

The leading labs, at their usual pace

Expected next

See the timeline

Estimated from each lab's usual gap between releases. The labs don't announce these dates.

Independent tests, in plain English

Best right now

Which AI should I use?All rankings

Best all-rounder

GPT-6 AstraOpenAI

GPT-6 Astra, GPT-6.1 Sol, Claude Opus 5.5 and GPT-6.1 Terra are effectively tied at the top of Epoch AI's combined test score: the gaps between them are within the margin of error.

Next: GPT-6.1 Sol · Claude Opus 5.5

People's favourite

Gemini 4 CarbonGoogle· not public yet

In blind votes, Gemini 4 Carbon and Gemini 4 Argon are neck and neck: Gemini 4 Carbon's answers win about 51% of the time.

Next: Gemini 4 Argon · Muse Wonder

Best for coding

Claude Opus 5.5Anthropic

Claude Opus 5.5 builds the apps people like best: in blind votes it beats Claude Sonnet 5.5 about 56% of the time.

Next: Claude Sonnet 5.5 · Gemini 4 Carbon

Best free to download

DeepSeek V4.1 FlashDeepSeek· Open

DeepSeek V4.1 Flash is the strongest model you can download and run yourself, scoring 159 against 169 for the best model overall.

Next: Kimi K3 · GLM-5.3

Channels

Alerts where you already are.

TelegramOne tap, no account DiscordJoin our server EmailLaunches, digest, weekly ntfyOpen-source push RSS & JSONFor your reader

Questions

Good to know.

I'm new to AI. Will I understand the alerts?

Yes. Every release starts with one sentence in plain English, and says where you can try it: in an app like ChatGPT, for developers, or as a free download. Words like “parameters” or “open weights” have a short definition you can hover over or tap, and the glossary explains them all.

How fast are the alerts?

trckr checks the labs' own systems, Hugging Face: The website where labs and researchers publish models anyone can download. Think of it as an app store for AI models. and OpenRouter: A service that gives developers one place to use hundreds of models from many labs. New models often appear there early, sometimes anonymously. every 15–60 seconds, and official blogs every minute or two. An alert usually reaches you within a minute of the first public trace of a model, often before the official announcement.

What counts as a new model?

New AI models from the major labs: the models behind assistants like ChatGPT, Claude and Gemini, free-to-download models, and image, video and audio models. Each release is rated Major, Notable or Minor, and you choose what you hear about. Copies and size variants of one launch arrive as one alert.

What are early signals?

Hints that a model is coming before it's announced: anonymous test models, model names that show up in a lab's systems before launch, and code that adds support for a new model. They're listed apart from the releases (under Spotted early) and are off by default in alerts, because they're occasionally wrong.

Do I need an account?

No for browser push, Telegram or the trckr Discord server: one tap and you're in. An account (email sign-in link, no password) lets you send alerts to your own Discord server, email or ntfy, and manage everything in one place.

Push on iPhone?

Yes, on iOS 16.4 or newer: add trckr to your Home Screen from Safari's Share menu, open it from the icon, and enable push there. Telegram works everywhere without that step.

What do you do with my data?

Only what's needed to deliver alerts: your email if you sign up, and the address of each destination you add. No ads, no trackers, nothing sold. Delete your account anytime and it's gone.

Stop refreshing.

Get a ping the moment a new model drops, wherever you are.

Free, no account needed. Push, Telegram, Discord, email or RSS.