I can’t use haiku-4.5 with the cursor cli (agent), and when I ask Cursor to use a subagent with it (which I assume it just calls agent) it also cannot use it. It is my preferred model for many cases due to its speed and reliability and cost. Surely, other models are also missing, but I am particularly missing this one.
Steps to Reproduce
Run agent models and it will be missing from the list:
$ agent models
Available models
auto - Auto
composer-1.5 - Composer 1.5
composer-1 - Composer 1
gpt-5.3-codex-low - GPT-5.3 Codex Low
gpt-5.3-codex-low-fast - GPT-5.3 Codex Low Fast
gpt-5.3-codex - GPT-5.3 Codex
gpt-5.3-codex-fast - GPT-5.3 Codex Fast
gpt-5.3-codex-high - GPT-5.3 Codex High
gpt-5.3-codex-high-fast - GPT-5.3 Codex High Fast
gpt-5.3-codex-xhigh - GPT-5.3 Codex Extra High
gpt-5.3-codex-xhigh-fast - GPT-5.3 Codex Extra High Fast
gpt-5.2 - GPT-5.2
gpt-5.3-codex-spark-preview - GPT-5.3 Codex Spark
gpt-5.2-codex-low - GPT-5.2 Codex Low
gpt-5.2-codex-low-fast - GPT-5.2 Codex Low Fast
gpt-5.2-codex - GPT-5.2 Codex
gpt-5.2-codex-fast - GPT-5.2 Codex Fast
gpt-5.2-codex-high - GPT-5.2 Codex High
gpt-5.2-codex-high-fast - GPT-5.2 Codex High Fast
gpt-5.2-codex-xhigh - GPT-5.2 Codex Extra High
gpt-5.2-codex-xhigh-fast - GPT-5.2 Codex Extra High Fast
gpt-5.1-codex-max - GPT-5.1 Codex Max
gpt-5.1-codex-max-high - GPT-5.1 Codex Max High
gpt-5.4-high - GPT-5.4 High
opus-4.6-thinking - Claude 4.6 Opus (Thinking) (default)
gpt-5.4-low - GPT-5.4 Low
gpt-5.4-medium - GPT-5.4
gpt-5.4-medium-fast - GPT-5.4 Fast
gpt-5.4-high-fast - GPT-5.4 High Fast
gpt-5.4-xhigh - GPT-5.4 Extra High
gpt-5.4-xhigh-fast - GPT-5.4 Extra High Fast
sonnet-4.6 - Claude 4.6 Sonnet
sonnet-4.6-thinking - Claude 4.6 Sonnet (Thinking) (current)
opus-4.6 - Claude 4.6 Opus
opus-4.5 - Claude 4.5 Opus
opus-4.5-thinking - Claude 4.5 Opus (Thinking)
gpt-5.2-high - GPT-5.2 High
gemini-3.1-pro - Gemini 3.1 Pro
grok - Grok
sonnet-4.5 - Claude 4.5 Sonnet
sonnet-4.5-thinking - Claude 4.5 Sonnet (Thinking)
gpt-5.1-low - GPT-5.1 Low
gpt-5.1 - GPT-5.1
gpt-5.1-high - GPT-5.1 High
gemini-3-pro - Gemini 3 Pro
gemini-3-flash - Gemini 3 Flash
gpt-5.1-codex-mini - GPT-5.1 Codex Mini
kimi-k2.5 - Kimi K2.5
Tip: use --model <id> (or /model <id> in interactive mode) to switch.
Haiku 4.5 (and a few other models) is currently intentionally unavailable in the CLI. This is a known limitation, not a bug. In the IDE, this model works as usual.
As a workaround for now, you can use Haiku through the IDE, or via subagents in the IDE.
The IDE refuses to use haiku-4.5 as a subagent. I have tried many times and it just does not, it falls back to composer-1.5 in all my tests (well, once it used the same model I was asking it from to use it as a subagent).
Subagents are a separate known issue. A few users reported that subagents ignore the selected model and fall back to composer-1.5 (similar thread). The team is aware.
A couple quick questions to clarify:
How exactly are you asking it to use haiku-4.5 for the subagent, via a custom .md file with model: haiku-4.5 in the frontmatter, or are you just telling the agent in chat “use haiku as subagent”?
Can you share the Request ID?
That’ll help us see what’s happening on the backend.
Hello Dean, here you have a request ID of a chat where I requested to use haiku-4.5-thinking and it used composer-1.5: 266c96c9-136b-497e-af58-6c27d70e7677
Here you have another one where not only it did the same: cbf38e49-0068-43db-9926-b8fe751b8df4
This one I also told it to NOT use opus-4.5 and it did use opus-4.5 as a subagent against my instructions.
The actual prompt piece said:
Always use subagents to do work. Use the model `haiku-4.5-thinking` for simple things (reading files/summarising/executing tools and reporting result), use `sonnet-4.6-thinking` for tasks that need reasoning and iterating. Do not use `opus-4.6`.
In this last one, the AGENTS.md also says the same. In the first one I’m not completely sure.
I reached my usage limit so I can’t really use Cursor as of now (and my team is confused as of what and where “limit for faster responses” is to increase it for me) to further test for you.
There has been a change in the system prompt or the tool for the subagents. It now reports it can only state “fast” (which goes to composer-1.5) to choose a model, or, it will use the model set in the chat.
Now my chats will only use those 2 models no matter what I ask cursor to do. This is pretty disappointing @deanrie
A few questions have piled up, so I’ll answer in order.
@SamP: About subagents and model selection. This is a known issue. Subagents don’t always inherit the model you set and can fall back to composer-1.5. A few users are seeing the same thing, similar thread: Nested subagents don't inherit model in MAX mode. For now, sadly, there isn’t a reliable workaround. Subagents will use either the model from the chat or the fast model composer-1.5. There’s no ETA for a fix, but reports help with prioritization.
@philm, @Mark_Lemay: About cheap and fast models like gpt-5.4 Mini, Nano, etc in the CLI. There’s no info on an ETA for adding them yet. Right now the model list in the CLI is intentionally limited, but your requests help when we review these decisions.
Thanks! Looks like they were added to the CLI late last week some time. Though they did change the name of the default gpt-5.4 to be gpt-5.4-medium which broke a bunch of our CI. not too big a deal but considering the updated name follows the pre-existing model naming pattern, they could’ve called it that from the get go instead of pulling an unannounced switcheroo.