Claude Haiku 4.5 is now available in Cursor. You can find more information on pricing here. Please try it out and let us know what you think!
I’m glad the pricing is so transparent these days.
By the way, the pricing information about Haiku 4.5 at your link you’ve mentioned is invisible to my eye, can’t find anything. I’ve clicked “Show more models“, yeap.
There is no information about prices, just dashes.
What are its capabilities? Context window, thinking?, image handling?, tool usage?
Claude Haiku 4.5 is available everywhere today. If you’re a developer, simply use claude-haiku-4-5 via the Claude API. Pricing is now $1/$5 per million input and output tokens.
Looks like 1/3 the cost of Sonnet-4.5 at slightly better capability than Sonnet-4.0…roughly. Will play with it to see if there are also token efficiency improvements to reduce the cost comparison further than just the retail price. Could have a place in my toolkit!
Tried it a few times and it could not use the edit tool. Tried sed, etc. The first task I gave it was to remove a try/catch and it could not figure out how to edit the file.
I’ll try again after an update….or two.
It’s 1 request.
No tool usage shown on models doc page. It did use git + write an .md for me, tho. It seems very mid, and not THAT fast.
Don’t see it in Claude Code yet, could be a good little code slave for plan mode, not sure.
agree its mid.
They generally want to get ahead in coding, but it would be nice if they had a model that was purely for coding purposes, like the Codex.
It seems that Haiku ignores global rules. Even despite its reduced price, it is unclear in what scenarios it should be used. Codex is suitable for code, and the free grok code is good for quick searches. What is Haiku needed for, since it is worse in every way?
Those benchmark are counterproductive at this point. Models get optimized on them, instead of actual usage. The latter is quite hard to benchmark.
hmmm:
- no tool usage: will it be able to create files and folders?
cannot disable reasoning ? i think i’ve seen a version without reasoning, am i wrong?- EDIT: the non thinking model was disabled inside the settings.
So, is Haiku 4.5 not able to edit existing code? It seems to lack tool usage, and as far as I understand editing an existing file is done through a tool. If that is the case, what is the point of including Haiku 4.5 in Cursor? Lacking tool usage, I am honestly not sure what it can offer. I guess pure research tasks? However with broken @Docs, @Web, etc. support in Cursor right now, you can’t really even do that effectively…
I guess Haiku might be a planning-only model, but it does not look like it could stand in as a 1/3rd cost alternative to Sonnet 4.0, as it simply isn’t capable without tool usage capabilities.
Haiku often fails to apply edit_tool correctly the first time, but generally works as a full-fledged Agent.
It seems Haiku-Thinking is more or same expensive as gpt-5-high XD
I’ll write about it in my guide if there’s anything noteworthy.
I poked around a bit and found a bunch of info that indicated the model couldn’t use tools. I couldn’t seem to get it to edit any file. It will generate code samples and stuff in the agent chat, but, it didn’t seem to be able to use terminal, edit files, mcp, etc.
I have Cursor 1.7.44 (user setup), Windows 11 and Agent Compass in rules.
But I’ve done less than ten prompts so far.

