57
GPT-6 Luna works acceptably in agents but still misses some bugs
Claude Haiku 5.5 vs GPT-6 Luna
Which model delivers useful results in real work: Claude Haiku 5.5 or GPT-6 Luna?
- Point change
- +5 pts
- Current state
- Haiku 5.5
- Status
- Running
More monitor statisticsLess monitor statistics
- Point change
- +5 pts
- Current state
- Haiku 5.5
- Status
- Running
Real-work usefulness
preevia.comClaude Haiku 5.575GPT-6 Luna55
Updated Nov 8Last 30 days · 3-day rolling
Where people stand
A sample of the public voices behind the data.
@thercemanI’m currently experimenting with agents swarm lead…
tested GPT 6 Luna, Haiku 5.5, DeepSeek v4.1 Flash
Haiku lost instantly - too much guardrails
Luna is ok but doesnt want to delegate th
@blueemi99I compared both GPT-6 Luna and Claude Haiku 5.5 on a minecraft prompt, and wow, the difference is huge.
Haiku's one isnt perfect, but feels a lot closer to minecraft, with everything workin
@JustRouzbehMost people think the big model does all the work in a chatbot. It doesn't.
Before it answers, a tiny model has already done 3 jobs
Followed the conversation. Someone asks "do you ship to 
@SidraMiconiClaude Haiku 5.5, GPT-6 Luna and Gemini 3.8 Flash read 40 real contracts and answered three questions you ask before signing: does it renew on its own, is there a notice clause to stop it, c
@vishalanton_I've spent most of the last 24 hours trying Mercury 3 and Deepseek V4.1 flash 1017 in browser agents, alongside claude haiku 5.5 and gpt-6 luna.
I used Stagehand and Pi. Deepseek ran throug
@DemetriusZhomirAsked Claude Haiku 5.5 to explain a diff.
It said the diff adds a feature.
With the same prompt, GPT-6 Luna said it's documentation describing the feature with QA evidence.
Guess which one
@AlexandarM21834GPT-6.1 Terra fixed all 20 bugs in my app for about $.06 a run
GPT-6.1 SOL did the same for $.22 and opus 5.5 for $.78
GPT-6 Luna and Haiku 5.5 cost less but missed bugs in every run
@ptr1337_1/5 I tested Haiku 5.5 in native Claude desktop Code vs GPT-6 Luna in Codex CLI. 24 synthetic Python/structured-text tasks, 2 repeats at Low and Max. 192 primary outcomes.
Task successes by
@sshayahmetSWE 2 Flash put together a full 3D voxel diorama for $0, and kept improving it using screenshot feedback.
Ran it against Claude Haiku 5.5, Gemini Flash 3.8, and GPT-6 Luna on a procedural mClaude Haiku 5.5MixedGPT-6 Luna
The views shaping the conversation
Recurring ideas, summarized across multiple conversations.
Claude Haiku 5.5MixedGPT-6 Luna
57
Haiku 5.5 struggles in agent workflows and bug-fixing despite lower cost
40
Results depend heavily on the specific task; no clear overall winner
These are collected experiences, not a survey of all users. Errors and post miscalculations are possible.
34% signal support · 1,452 conversations