GPT · Last 7 days
GPT-6 Luna
0–100 · higher means more positive community experience, not benchmark performance.
Updated 2026-10-01 11:07 UTC · 2026-09-24 – 2026-10-01 UTCExperience score35.1/100112 comments in 7 days
What people think
One user notes that on the DeepSWE v1.1 software engineering test, GPT-6 Luna scored 66.6%, near Opus 5/Fable 5 medium effort, with per-task cost ~93%/96% lower. One user asked GPT-6 Luna to use computer use to solve a problem, but it stubbornly refused despite repeated reminders
About the sample and scores
As of 2026-10-01 11:27:03 UTC, GPT-6 Luna in the GPT family has 112 explicitly attributed community comments in the last 7 days; its community score is 35.1/100. Available category scores include General text 37.4/100 (n=51); Coding 34.9/100 (n=14); Reasoning 36.6/100 (n=15). Scoring method and sources
Community reviews
Selected comments from the last 7 days. The balance of excerpts does not represent the share of positive reviews.
2 selected excerpts
Community reports about limits cannot establish official allowances.
Show original
GPT-6 Luna 得分为 66.6%，接近 Opus 5 和 Fable 5 的 medium effort，单任务成本则低约 93% 和 96%
Comment context
Software engineering benchmarks tell a clearer story. In DeepSWE v1.1, GPT-6 Sol's max effort score is 68.8%, just 1.1 percentage points behind Claude Fable 5's top score of 69.9%, but with a per-task cost about 80% lower. GPT-6 Luna scores 66.6%, close to Opus 5 and Fable 5 at medium effort, with per-task costs about 93% and 96% lower respectively.
Machine translatedShow original
软件工程测试更能说明问题。在 DeepSWE v1.1 中，GPT-6 Sol 的 max effort 得分为 68.8%，距离 Claude Fable 5 的最高成绩 69.9% 只差 1.1 个百分点，但单任务成本低约 80%。GPT-6 Luna 得分为 66.6%，接近 Opus 5 和 Fable 5 的 medium effort，单任务成本则低约 93% 和 96%。
Show original
叫它用computeruse帮我解决问题，提醒半天死活不敢
Show original
6luna是个又蠢又慢的模型
Comment context
I've had bad experiences with gpt 6 luna, but 5.6 luna worked real nicely with e.g React, you can do a back n forth and it'll work like a teammate vs just a "done!" where you then have to clean everything up afterwards
Show original
要开最高档，反复沟通两三轮才能出一点东西
Show original
速度貌似是比5.6时候快了一些
Show original
能力上明确强于DeepSeekv4f和pro
Show original
稍差一点，但是我觉得当子智能体的话不耽误啊，改一下主智能体的提示词，让它把复杂任务拆细就是了。
Comment context
I haven't used it much recently. Is 6 luna a regression compared to 5.6 luna? I previously set the sub-agent to force 6 luna max; do I need to change it back to 5.6 luna max? [crying laughing]
Machine translatedShow original
我最近没怎么用，6 luna跟5.6luna比难道是退步的么？我之前是设置了子代理强制用6luna max，需要改回5.6 luna max么[笑哭]
Show original
我看没什么区别啊
Comment context
Is there any substantive improvement in 6luna compared to 5.6luna? I don't see any difference.
Machine translatedShow original
6luna和5.6luna有什么实质性提升吗？我看没什么区别啊
Show original
特别推荐GPT6 luna，量大管饱，性能跟sol差距不大
No selected comments for this filter in the last 7 days.