How to follow AI model releases without drowning
New models ship every week. How to tell which releases matter to you, and where to watch for them.
18 Oct 2026
There are more AI model releases than anyone can track: new open-weight models, new versions in APIs, new benchmark claims. Most won't change your work. Here is how to find the few that do.
Decide what you actually care about
Ask yourself which of these you'd act on:
- A better model in a product you use (a chat assistant, a coding tool). You'll notice it anyway; a headline is enough.
- A cheaper or faster model in an API you pay for. Worth reading: it can cut costs.
- An open-weight model you could run yourself. Worth reading if you self-host or care about privacy.
- A research result. Interesting, rarely urgent.
Everything outside your list can be a headline you skim.
Where releases are announced first
Companies announce on their own blogs and model pages before anyone writes about them. ReadKernel reads the main ones directly (OpenAI, Google DeepMind and Google's AI blog, NVIDIA, Hugging Face), plus outlets that cover releases carefully. See the full list on our sources page.
How to judge a release in two minutes
- What's new, in one sentence? If the announcement can't say it plainly, be sceptical.
- Benchmarks: whose numbers? A company's own benchmark chart is a claim, not a result. Wait for independent tests on tasks like yours.
- Price and limits. Cost per million tokens, context length, rate limits. This is often the real news.
- Licence. "Open" ranges from fully open to "weights available with conditions". Read it before you build on it.
Let ReadKernel do the watching
ReadKernel tags stories about new or updated models as Model releases and shows them in their own box at the top of the AI feed. If AI is all you want, build a brief with only the AI topic.