Release Notes: All Three Extensions Now Run on Claude Sonnet 5

All three Comment Generator extensions now run on Claude Sonnet 5. Instagram, LinkedIn, and Threads switched on the same day, so whichever platform you comment on, the drafts you get from today onward come from a newer model.

Short version: comments read more like something you'd type yourself. Sonnet 5 is the best model we've tested at picking up the register of a post and answering in it, which is the exact thing older models kept getting wrong. Nothing else moved. Same buttons, same settings, same rule that you read the draft before it goes anywhere.

Which extensions were updated?

All of them. Comment Generator for Instagram, Comment Generator for LinkedIn, and Comment Generator for Threads are on Sonnet 5 as of 2026-08-07. The Comment Generator API runs the same model, so anything built on top of it picked up the change at the same time.

Why Claude Sonnet 5

A comment is a small piece of writing where every word is visible. There's no length to hide behind and no structure to carry it. When a weaker model writes one, the problem usually isn't a factual mistake. It's a sentence that answers the post correctly and still sounds like a machine produced it: too balanced, too polished, three clauses where a person would have written one.

That's the part Sonnet 5 handles better. We ran it against the model it replaced on real captions from all three platforms before switching, and the drafts came back needing less editing. Anthropic positions it as the strongest model in the Sonnet line for natural language, and for writing comments that's the benchmark that actually matters.

It matches the tone of a post instead of averaging it

A dry industry post on LinkedIn and a one-line joke on Threads used to come back with answers of a similar shape. Sonnet 5 reads the register more accurately, so a casual caption gets a casual reply and a technical one gets something with substance in it. Your tone setting still wins whenever you've picked one. The model just starts from a better guess when you leave it to decide.

It reads more of the post

Sonnet 5 works with a much larger context window than the model it replaced. In practice that means long LinkedIn posts, and Threads conversations with several replies stacked above the one you're answering, get read properly instead of mostly from the opening lines. Comments on long posts should feel noticeably less generic as a result.

Comments in other languages improved

Auto-detect was already there, but quality varied by language. German, French, Spanish, Portuguese, Indonesian, Hindi, Chinese, and Urdu all gain from this change, and the gap between an English draft and a non-English one is smaller than it was. If you set a preferred language in the settings, that still overrides detection.

Do you need to do anything?

No. The model runs on our side rather than inside the extension, so there's no update to install and no version number to check. Open Instagram, LinkedIn, or Threads and the next comment you generate already uses Sonnet 5. Your settings came across untouched: language, length, formality, the end-with-a-question toggle, your closing line, and any instructions you've saved in your own words.

What did not change

You still review every draft. The comment lands in the platform's native comment box and stays there until you post it yourself, exactly as before. A better model makes that step feel more skippable, which is precisely why it's worth naming: comments get flagged as spam because nobody read them before they went out, not because the model wasn't good enough. We went through that argument in why AI-drafted comments still need a human in the loop, and a model upgrade doesn't change any of it.

Pricing stays where it was. The 7-day free trial still doesn't ask for a card, and existing plans keep the limits they had.

What's next

We're watching how Sonnet 5 behaves on each platform separately, because Instagram, LinkedIn, and Threads reward fairly different things from a comment. Where the defaults are off, we'd rather tune the prompt per platform than leave you to patch it in your own instructions. If a draft comes out wrong in a way that looks systematic, tell us which platform it was and roughly what the post was about. Those reports are what shape the next round of tuning.