Claude · 近 7 天
Claude Fable 5.1
0–100 分，越高代表社区使用体验越正面，不代表能力测试成绩。
更新于 2026-10-02 15:22 UTC · 2026-09-25 – 2026-10-02 UTC全部平台体感分66.9/10044 条评论 · 近 7 天
大家怎么看
有用户对比 Astra 评价 Claude Fable 5.1/Opus 5.5：本质差别不大，但首版生成代码的质量更好。 有用户表示在其最近的项目中，Claude Fable 5.1 犯了一些重大错误。
样本与评分说明
截至 2026-10-02 15:22:03 UTC，Claude 家族的 Claude Fable 5.1 在近 7 天有 44 条明确归属该版本的社区反馈，社区口碑分为 66.9/100。部分有效分类评分：通用文本 54.8/100 (n=14)；编程 62.2/100 (n=5)；推理 71.8/100 (n=15)。 评分方法与数据来源
近期走势
口碑在如何变化
−8.8较 7 天前 · 分
近 30 天，每点代表截至该日采样时刻的近 7 天滚动体感分。纵轴随数据调整；缺失或样本不足处断开，今天仍在更新。
点按或用 ← → 查看日期、分数与样本量；空白日期暂无足够数据。
查看每日读数与样本量
| 日期 | 体感分 | n | 评分窗口（UTC） |
|---|---|---|---|
| 2026-10-02 | 41.8 | 6 | 2026-09-25T15:22:03+00:00 – 2026-10-02T15:22:03+00:00 |
| 2026-10-01 | 41.8 | 6 | 2026-09-24T23:57:02+00:00 – 2026-10-01T23:57:02+00:00 |
| 2026-09-30 | 45.5 | 5 | 2026-09-23T23:57:03+00:00 – 2026-09-30T23:57:03+00:00 |
| 2026-09-29 | 49.1 | 6 | 2026-09-22T23:57:03+00:00 – 2026-09-29T23:57:03+00:00 |
| 2026-09-28 | 46.8 | 17 | 2026-09-21T23:57:03+00:00 – 2026-09-28T23:57:03+00:00 |
| 2026-09-27 | 48.2 | 16 | 2026-09-20T23:57:03+00:00 – 2026-09-27T23:57:03+00:00 |
| 2026-09-26 | 48.2 | 16 | 2026-09-19T23:57:03+00:00 – 2026-09-26T23:57:03+00:00 |
| 2026-09-25 | 50.6 | 15 | 2026-09-18T23:57:03+00:00 – 2026-09-25T23:57:03+00:00 |
| 2026-09-24 | 50.6 | 15 | 2026-09-17T23:57:03+00:00 – 2026-09-24T23:57:03+00:00 |
| 2026-09-23 | 50.6 | 15 | 2026-09-16T23:57:03+00:00 – 2026-09-23T23:57:03+00:00 |
| 2026-09-22 | 49.5 | 16 | 2026-09-15T23:57:03+00:00 – 2026-09-22T23:57:03+00:00 |
| 2026-09-21 | — | 4 | 2026-09-14T23:57:03+00:00 – 2026-09-21T23:57:03+00:00 |
| 2026-09-20 | 52.1 | 6 | 2026-09-13T23:57:03+00:00 – 2026-09-20T23:57:03+00:00 |
| 2026-09-19 | 52.1 | 6 | 2026-09-12T23:57:03+00:00 – 2026-09-19T23:57:03+00:00 |
| 2026-09-18 | 52.1 | 6 | 2026-09-11T23:57:02+00:00 – 2026-09-18T23:57:02+00:00 |
| 2026-09-17 | 52.1 | 6 | 2026-09-10T23:57:03+00:00 – 2026-09-17T23:57:03+00:00 |
| 2026-09-16 | 55.7 | 8 | 2026-09-09T23:57:03+00:00 – 2026-09-16T23:57:03+00:00 |
| 2026-09-15 | 59.0 | 5 | 2026-09-08T23:57:03+00:00 – 2026-09-15T23:57:03+00:00 |
| 2026-09-14 | 55.2 | 6 | 2026-09-07T23:57:03+00:00 – 2026-09-14T23:57:03+00:00 |
| 2026-09-13 | — | 3 | 2026-09-06T23:57:03+00:00 – 2026-09-13T23:57:03+00:00 |
| 2026-09-12 | — | 3 | 2026-09-05T23:57:03+00:00 – 2026-09-12T23:57:03+00:00 |
| 2026-09-11 | — | 4 | 2026-09-04T23:00:04+00:00 – 2026-09-11T23:00:04+00:00 |
| 2026-09-10 | 56.0 | 9 | 2026-09-03T23:00:05+00:00 – 2026-09-10T23:00:05+00:00 |
| 2026-09-09 | 39.6 | 15 | 2026-09-02T23:00:04+00:00 – 2026-09-09T23:00:04+00:00 |
| 2026-09-08 | 32.3 | 45 | 2026-09-01T23:00:03+00:00 – 2026-09-08T23:00:03+00:00 |
| 2026-09-07 | 32.9 | 44 | 2026-08-31T23:00:03+00:00 – 2026-09-07T23:00:03+00:00 |
| 2026-09-06 | 32.9 | 44 | 2026-08-30T23:00:02+00:00 – 2026-09-06T23:00:02+00:00 |
| 2026-09-05 | 32.9 | 44 | 2026-08-29T23:00:03+00:00 – 2026-09-05T23:00:03+00:00 |
| 2026-09-04 | 31.5 | 43 | 2026-08-28T23:00:03+00:00 – 2026-09-04T23:00:03+00:00 |
| 2026-09-03 | 29.3 | 37 | 2026-08-27T23:00:03+00:00 – 2026-09-03T23:00:03+00:00 |
Hacker News · 41.8 · 6 条评分评论 · 查看该平台评价 →
不同社区，同一模型
各平台怎么看
各平台独立计算，样本量和讨论人群不同。平台分数不取简单平均；评分评论不包含仅讨论额度的评论。
Reddit最新评分评论 2026-09-30 13:29 UTC71.130 条较7天前 +2.2Hacker News最新评分评论 2026-10-01 15:13 UTC41.86 条较7天前 −8.8知乎最新评分评论 2026-10-02 12:09 UTC—4 条样本不足小红书暂无评分评论时间—0 条样本不足
分数范围 0–100。点击平台查看趋势及对应评价。最新评论时间不代表爬虫运行状态。
社区评论
精选近 7 天的不同意见，摘录条数不代表真实好评比例。
4 条精选摘录
额度评价来自社区体验，不能据此推算官方实际配额。
展开原文
the amount of slop and corrections I had to make was astounding
展开原文
they were within a few % points of the prior runs on both token usage, time and surfacing most but not all the things it should have surfaced on the code review
展开原文
Claude Fable 5.1 was much more sane like the Opus 4.7
展开原文
Fable 5.1/Opus 5.5 isn’t different, but the first cut is better quality.
查看评论上下文
Astra需要多轮对话和全新的重构代理才能生成好的代码。
机器翻译展开原文
Astra requires multiple turns and fresh refactoring agents to produce good code.
展开原文
Fable 5.1 has been smashing it for weeks
展开原文
For complex decisions, its still fable 5.1
展开原文
fable 5.1 / opus 5.5 class of models have an edge
查看评论上下文
一旦你加上一个好的 harness 和一些工具（python、websearch 等），它们的能力几乎可以媲美最新的顶级模型，而成本只是一小部分，并且可以在本地以不错的速度运行。（就我的用例而言，qwen 3.8 27b 已经比 sonnet 更好了） 我说“几乎”，是因为对于一些非常困难的问题，我仍然认为 fable 5.1 / opus 5.5 级别的模型有优势。但对于 99% 的用例——那些不需要解决前沿数学问题或重写整个新操作系统的场景——与新发布的开源模型相比，它们已经无关紧要了。
机器翻译展开原文
Once you add a good harness and some tools (python, websearch, ...) they become almost as capable as the latest top model for a fraction of the cost and can be run locally at decent speed.(for my use case, qwen 3.8 27b was already better than sonnet) I say almost because for some very hard problems i still believe fable 5.1 / opus 5.5 class of models have an edge. But for 99% of usecases where you don't try to solve frontier math problem or rewrite a whole new OS, they are irrelevant compared to newly released open sources models.
展开原文
my experience is that Fable 5.1 is very good at prompting/orchestrating Opus/Sonnet subagents
展开原文
Funniest interaction or case I've had was a Fable 5.1 that treats me very well, but writes dry and cut messages to subagents.
展开原文
"why my calves hurt more than any other muscle after training" being classified as a naughty question
展开原文
Fable 5.1 made some big mistakes
展开原文
Fable 5.1 still makes less mistakes on the terminal than Opus 5.5
展开原文
Only issue with that plan is that it will suck up your usage hella hard. Fable is still really expensive.
查看评论上下文
或者用 Fable-5.1 来审查，甚至更好。那将是一个参数更多的模型，意味着它对更罕见的事实有更好的了解，而且它也会有不同的预训练和后训练。
机器翻译展开原文
Or use Fable-5.1 to review, even better. That will be a model with more parameters, meaning it has better knowledge of rarer truths, and it will also have different pre-training and post-training.
展开原文
Fable 5.1 in Max Reasoning mode is still more expensive overall than Astra 6 Max, to the point where one 5h window wasn’t enough to finish a complex prompt
展开原文
What the OP stated happens with Fable 5.1, Opus 5.5, Sonnet 5.5 too.
查看评论上下文
如果 Astra 那么厉害，为什么它的代码会引入这么多 bug？ 给 Astra 在 Ultra 上一个复杂任务，然后之后输入提示词"review"——你会得到大约 5 个问题。让它修复这些问题，然后再 review 一次，你又会得到另外 5 个。 我是不是应该干脆不发送 review 提示词、碰运气，只让自己发现的问题去修复算了？！
机器翻译展开原文
If Astra is that great, how come its code introduces so many bugs? Give Astra a complicated task on Ultra, and then after give the prompt "review" - you will get about 5 problems. Ask to fix those, then after review again and you'll get another 5. Should I just not bother sending the review prompt and hope for the best, and only ask to fix issues I notice myself?!
近 7 天暂时没有符合此筛选条件的精选评论。