Fable

188 atoms · 跨 23 天 · 首见 2026-06-09 · 最近 2026-07-03

三色: 🟦 fact 59 · 🟥 take 129 · stance ▲80/▼33/◆31

叙事状态: 🍂 fading (退潮) · 💤 沉寂 · 策展近14d 0 atom · 🐦 X 近14d 31

状态只算低频策展源 (群/卖方); X firehose 仅作背景音量

来源: X 188

时态: fresh:173 · stale:9 · aging:6

标签: 好观点:117 · 好数字:38 · 好思考:31 · 好信源:16 · 好问题:6

🟨 AI 综合 · junior analyst 概览

🧭 拥挤度 (一人一票): ▲ 30 位作者 (KOL30) vs ▼ 25 位作者 (KOL25)

⚖️ 多头 30/30 来自KOL

💢 核心分歧 (3 轴)

1. 模型真实能力 vs 华而不实

*topic: 技术路线/模型性能/模型训练/监管政策 · 41 bull vs 14 bear*

🟢 bullish 侧:

• <a id="atom-4f716b0bd456afbd"></a>2026-06-09 X·@emollick · capability jump · 模型发布/技术路线 +同日1条
  A genuine jump in capability, I could feed it a 15 page design document for a project and it would work for 9+ hours and deliver terrific results.
  tags: 好观点 · → daily
• <a id="atom-1a8b7807e7842835"></a>2026-06-10 X·@rahulchhabra07 · 情绪智能评价 · 技术路线
  a lot of emotional intelligence too
  tags: 好观点 · → daily
  value: direction=a lot of emotional intelligence too
• <a id="atom-31bc9f4827cba193"></a>2026-06-10 X·@blader · 智能飞跃评价 · 技术路线 +同日2条
  the intelligence leap is pretty crazy though
  tags: 好观点 · → daily
  value: direction=pretty crazy
• <a id="atom-36a9631c6d114f82"></a>2026-06-10 X·@scaling01 [老旧] · FrontierCode 性能 · 模型性能
  it destroys all other models
  tags: 好观点 · → daily
  value: direction=destroy all other models
• <a id="atom-91237d5961bf89c7"></a>2026-06-11 X·@levie · Box AI Complex Work Eval score Media & Entertainment · 模型性能/垂直行业 +同日9条
  Media & Entertainment (78% vs 61%)
  tags: 好数字 · → daily
  value: qty=78% · comparison=vs Opus 4.8 61%

🔴 bearish 侧:

• <a id="atom-143548c0204b394b"></a>2026-06-10 X·@Teknium · is blocked from life sciences broadly · 监管政策/技术路线
  Fable is blocked from life sciences broadly, nerfed even if you get passed the classifiers and filter level blocks.
  tags: 好思考 · → daily
  value: direction=nerfed
• <a id="atom-e8a2134be6908ba5"></a>2026-06-10 X·@RYANHINGSHING [老旧] · 模型质量比Mytho4.6差 · 产品发布/技术路线 +同日1条
  弱智版mytho，比4.6還差
  tags: 好观点 · → daily
• <a id="atom-dea040ddb140091a"></a>2026-06-10 X·@ShanghaoJin [老旧] · 耗token比4.6快 · 技术路线
  耗token快很多
  tags: 好观点 · → daily
• <a id="atom-0a48e22b7226a3da"></a>2026-06-10 X·@tensecorrection [老旧] · 不可信任用于生成计划之外的任何用途 · 技术路线/产品发布
  don't even trust it for that
  tags: 好观点 · → daily
• <a id="atom-2d8465146739062e"></a>2026-06-11 X·@xeophon · 回复质量不均匀 · 模型质量/技术路线
  spent the day with fable on a bunch of random stuff and it’s very spiky, imo. it is brilliant in the same message where it makes 1-3 severe mistakes, which means you have to check even more stuff more in-depth, wasting time and tokens
  tags: 好观点 · → daily

2. 产品定价与用户价值匹配

*topic: 价格动态/定价权 · 3 bull vs 3 bear*

🟢 bullish 侧:

• <a id="atom-04bdf802b16f7095"></a>2026-06-15 X·@8teAPi · 成本性能比 · 价格动态/竞争格局 +同日1条
  Fable often comes in strictly cheaper than Opus.
  tags: 好观点 · → daily
• <a id="atom-f608b56e8eab8839"></a>2026-07-03 X·@downingARK · 成本对比 GLM 5.2 · 定价权
  Fable (low effort) is 7x cheaper than open-source GLM 5.2 (high effort) at the same accuracy
  tags: 好数字 · → daily
  value: qty=7x cheaper · date=2026-07-03

🔴 bearish 侧:

• <a id="atom-2d4bd6132f254aa7"></a>2026-06-10 X·@RYANHINGSHING [老旧] · 定价贵 · 定价权
  還收那麼貴
  tags: 好观点 · → daily
• <a id="atom-e9a52b2b38f26201"></a>2026-06-10 X·@ShanghaoJin [老旧] · API定价高于竞争对手 · 定价权
  不用max plan调API已经比人贵了
  tags: 好观点 · → daily
• <a id="atom-e25a1cdaaa392cc4"></a>2026-06-11 X·@ShanghaoJin · 用户使用成本 · 价格动态
  我今天早上开fable做了4个进程，半小时就超过max plan了。然后就是按照API架构收费了
  tags: 好数字 · → daily

3. 创新独特性还是数据污染幻觉

*topic: 技术创新/模型能力 · 7 bull vs 1 bear*

🟢 bullish 侧:

• <a id="atom-5f075b8e1b4769fe"></a>2026-06-12 X·@_ueaj · invents new terminology · 技术创新_ +同日2条
  Fable loves to make up new words/terminology
  tags: 好观点 · → daily
• <a id="atom-cbe6b1eea8786e56"></a>2026-07-01 X·@theo · 能跑其他 agent · 模型能力 +同日2条
  Running other agents. It's the first model that feels like it actually gets how to use and orchestrate agents.
  tags: 好观点 · → daily
• <a id="atom-ea54e681dee8bc97"></a>2026-07-02 X·@rronak_ · 系统设计可靠性 · 模型能力
  Fable is the first model where I've started to think of it as a coworker, and where I actually have trust in its system design.
  tags: 好观点 · → daily
  value: direction=高

🔴 bearish 侧:

• <a id="atom-d8ead1cb192e1ccd"></a>2026-06-11 X·@usr_bin_roygbiv · 拒绝终端基准测试 · 模型能力
  Fable refuses nearly any terminal bench task.
  tags: 好观点 · → daily
  value: qty=nearly any · date=2026-06-11
  📷 原图

🔗 因果传导 (causal map)

*Fable 在产业链上的传导关系. 边是群里/卖方陈述的因果 (非 AI 推断), 数字 = 几条 atom 支撑. 点 atom 溯源.*

↓ 下游·近期陈述 (1)

• 🟢 US startups · 产能依赖 · 1 次陈述
  US startups need Fable and 5.6 to stay competitive.

↑ 上游·近期陈述 (1)

• 🟢 export controls · 政策影响 · 1 次陈述
  if not for circumstances around export control, Fable would of been terrible for market share

🟦 客观事实 (facts) (30)

• <a id="atom-ce208a9e8a3b3e07"></a>🟦 2026-06-18 X·@teortaxesTex · 目标发布时间为 2027 年 Q1 · 产品发布/路线图
  Q1 2027 is their target for Fable class
  tags: 好数字·好信源 · → daily
  value: date=Q1 2027
• <a id="atom-331868da761a4199"></a>🟦 2026-06-10 X·@blader · 前端内存压力优化效果 · 产品发布/技术路线
  it one shotted a 33% reduction in frontend memory pressure that no other human or agent found after a year of working on it
  tags: 好数字·好信源 · → daily
  value: qty=33% · direction=reduction
• <a id="atom-c2d85573f1c98210"></a>🟦 2026-06-26 X·@altcap · 出口管制状态 · 出口管制
  Fable is currently export controlled
  tags: 好信源·好观点 · → daily
  value: direction=管制中
• <a id="atom-b7cfd66acd994d9f"></a>🟦 2026-07-01 X·@brada · 产品已恢复可用 · 产品发布
  Fable is back
  tags: 好数字 · → daily
  📷 原图
• <a id="atom-ff7742b0a920e265"></a>🟦 2026-06-15 X·@TheTranscript_ · overreliance risk · 技术路线/竞争格局
  The situation we’re in collectively right now with Mythos and Fable is something that can happen with overreliance on certain models.
  tags: 好观点·好思考 · → daily
• <a id="atom-bb1dcc01c91c54a8"></a>🟦 2026-06-15 X·@TheTranscript_ · lesson acceptance · 技术路线/竞争格局
  But we will have done something wrong if we just accept this, don’t take the lesson, don’t build out and diversify.
  tags: 好观点·好思考 · → daily
• <a id="atom-f608b56e8eab8839"></a>🟦 2026-07-03 X·@downingARK · 成本对比 GLM 5.2 · 定价权
  Fable (low effort) is 7x cheaper than open-source GLM 5.2 (high effort) at the same accuracy
  tags: 好数字 · → daily
  value: qty=7x cheaper · date=2026-07-03
• <a id="atom-01d7caa792d62625"></a>🟦 2026-07-03 X·@sholtodouglas · produced PRs for @mainframe overnight · 产品发布_
  fable produced 60+ (!) ready to merge prs for @mainframe overnight.
  tags: 好数字 · → daily
  value: qty=60+
• <a id="atom-ac4bf5ca015fcaef"></a>🟦 2026-07-03 X·@sholtodouglas · user landed PRs in one day · 产品发布_
  I've landed like 12 PRs today this is fucking awesome
  tags: 好数字 · → daily
  value: qty=12 · date=today
• <a id="atom-d6aca6ba0107c058"></a>🟦 2026-07-02 X·@dejavucoder · price per 1M tokens · 价格动态
  fable 5 costs 10/50
  tags: 好数字 · → daily
  value: qty=$10/$50
• <a id="atom-2f6b762729dc8466"></a>🟦 2026-07-01 X·@PhiloGroves · 订阅价格 · 价格动态
  People are spending $200 for half a week of Fable, insanity.
  tags: 好数字 · → daily
  value: qty=$200 · period=half a week
• <a id="atom-10bc9f85d27acf79"></a>🟦 2026-07-01 X·@ChaseBrowe32432 · 订阅与API成本倍数关系 · 价格动态
  you can burn ~20,000$ a month worth of fable credits on the 200$ sub. it's literally 100x cheaper than API
  tags: 好数字 · → daily
  value: qty=100x
• <a id="atom-3a3503fd7dfd870f"></a>🟦 2026-07-01 X·@ChaseBrowe32432 · 订阅优惠期限 · 定价策略
  they're revoking it after a week which makes this impossible unfortunately
  tags: 好数字 · → daily
  value: date=after a week
• <a id="atom-e7d0938f181dfb1b"></a>🟦 2026-07-01 X·@thayto_dev · 解題時間 · 推理效率
  Fable 5 resolveu em 23m 53s
  tags: 好数字 · → daily
  value: qty=23m53s · date=2026-07-01
  📷 原图
• <a id="atom-34e889bf4702fc68"></a>🟦 2026-06-30 X·@SophiaCai99 · 出口管制状态预期 · 出口管制
  Commerce is expected to lift export controls on Fable tonight
  tags: 好信源 · → daily
• <a id="atom-b776bbf7e2131e1c"></a>🟦 2026-06-22 X·@teortaxesTex · 推理速度 · 模型发布
  Seeing as Fable was served at like 90 t/s
  tags: 好数字 · → daily
  value: qty=90 t/s · date= · direction=
• <a id="atom-321b901431351b6a"></a>🟦 2026-06-17 X·@teortaxesTex · 参数量 · 参数规模
  Fable is about 10T afaik
  tags: 好数字 · → daily
  value: qty=~10T
• <a id="atom-91237d5961bf89c7"></a>🟦 2026-06-11 X·@levie · Box AI Complex Work Eval score Media & Entertainment · 模型性能/垂直行业
  Media & Entertainment (78% vs 61%)
  tags: 好数字 · → daily
  value: qty=78% · comparison=vs Opus 4.8 61%
• <a id="atom-afe07b6b4d3473d4"></a>🟦 2026-06-11 X·@levie · Box AI Complex Work Eval score Technology · 模型性能/垂直行业
  Technology (81% vs 73%)
  tags: 好数字 · → daily
  value: qty=81% · comparison=vs Opus 4.8 73%
• <a id="atom-bac152d31f39fe2b"></a>🟦 2026-06-11 X·@levie · Box AI Complex Work Eval score Financial Services · 模型性能/垂直行业
  Financial Services (89% vs 83%)
  tags: 好数字 · → daily
  value: qty=89% · comparison=vs Opus 4.8 83%
• <a id="atom-572d3ab82be4d08d"></a>🟦 2026-06-11 X·@levie · Box AI Complex Work Eval score Healthcare · 模型性能/垂直行业
  Healthcare (66% vs 60%)
  tags: 好数字 · → daily
  value: qty=66% · comparison=vs Opus 4.8 60%
• <a id="atom-a34d26c5c02b73f6"></a>🟦 2026-06-11 X·@levie · Legal M&A due diligence eval score · 模型性能/合同审查
  Fable scored 100% vs Opus 4.8 78% on NDA term review
  tags: 好数字 · → daily
  value: qty=100% · comparison=vs Opus 4.8 78%
• <a id="atom-962d618b34541393"></a>🟦 2026-06-11 X·@levie · Healthcare clinical radiology error audit eval score · 模型性能/医疗诊断
  Fable 63% vs Opus 41% on radiology error categorization
  tags: 好数字 · → daily
  value: qty=63% · comparison=vs Opus 4.8 41%
• <a id="atom-9e5909d6e967025c"></a>🟦 2026-06-11 X·@levie · Media & Entertainment genre profitability projection eval score · 模型性能/财务建模
  Fable 74% vs Opus negative on genre profitability projection
  tags: 好数字 · → daily
  value: qty=74% · comparison=vs Opus 4.8 negative
• <a id="atom-c9e224c24e423088"></a>🟦 2026-06-11 X·@levie · Retail analytics eval score · 模型性能/数据分析
  Fable 94% vs Opus 61% on growth article analysis
  tags: 好数字 · → daily
  value: qty=94% · comparison=vs Opus 4.8 61%
• <a id="atom-d6f88ba92c6f6ad7"></a>🟦 2026-06-11 X·@levie · Financial Services 5-year debt facility projection eval score · 模型性能/财务建模
  Fable 83% vs Opus 62% on debt facility projection
  tags: 好数字 · → daily
  value: qty=83% · comparison=vs Opus 4.8 62%
• <a id="atom-0791b910a6a9e5c1"></a>🟦 2026-06-11 X·@levie · Technology SaaS feature valuation eval score · 模型性能/软件估值
  Fable scored 100% vs Opus 74% on Feature Value Index calculation
  tags: 好数字 · → daily
  value: qty=100% · comparison=vs Opus 4.8 74%
• <a id="atom-572008e56ee9fcc9"></a>🟦 2026-06-10 X·@jsnnsa · bugs solved · 软件交付
  they have solved 362 bugs today
  tags: 好数字 · → daily
  value: qty=362
  📷 原图
• <a id="atom-f1477b7d4ec5bfcf"></a>🟦 2026-06-10 X·@ticker2021 · context窗口设计为1m · 技术路线
  虽然写的是1m context
  tags: 好数字 · → daily
  value: qty=1M
• <a id="atom-455ab1507c9ade95"></a>🟦 2026-06-10 X·@mitchellh · price · 资本开支
  it took 2 hours and $40 to do it
  tags: 好数字 · → daily
  value: qty=$40

🟥 多头 takes (bullish) (30)

• <a id="atom-02774c3a141f1869"></a>🟥 2026-06-15 X·@scaling01 ⭐ · ECI score forecast · 模型发布/技术路线
  I suspect as we get more benchmark results its ECI should improve to ~163
  tags: 好观点·好思考 · → daily
  value: qty=~163 · date=future
• <a id="atom-70a81f87ca913f05"></a>🟥 2026-06-15 X·@scaling01 ⭐ · math performance improvement relative to prior Anthropic models · 技术路线
  relative to prior Anthropic models it also performs really well on math
  tags: 好观点·好思考 · → daily
  value: direction=better
• <a id="atom-caea3280888e0536"></a>🟥 2026-06-12 X·@scaling01 · 月度营收 · 营收预测
  Fable would have probably been something like ~$2B/month
  tags: 好数字 · → daily
  value: qty=~$2B/month · date=未明确时间
• <a id="atom-569cda9d08fbf1e6"></a>🟥 2026-06-10 X·@blader · SRE自动化性能 · 产品发布/技术路线
  automated oncall resolution - it is an extremely capable and reliable SRE. i've had an automation trying to do this before but it was never very good, now it's at or above human performance
  tags: 好观点·好信源 · → daily
  value: direction=at or above human performance
• <a id="atom-a26db3d3414cc668"></a>🟥 2026-07-01 X·@levie · 合作结果进展 · 合作客户
  Things seem to be ending up in a better spot with Fable
  tags: 好观点·好思考 · → daily
  value: direction=improving
  📷 原图
• <a id="atom-963b361e7e265436"></a>🟥 2026-06-18 X·@teortaxesTex · 能力预测 · 产品发布
  Fable or Fable+ could write *unbreakable* software.
  tags: 好观点·好思考 · → daily
• <a id="atom-7dd93abb7d33dad9"></a>🟥 2026-06-15 X·@krugermacro · performance comparison with frontier models · 竞争格局/模型性能
  Networks of smaller AI models are outperforming frontier models
  tags: 好观点·好思考 · → daily
  value: direction=outperforming
• <a id="atom-74c088e1c0431c0f"></a>🟥 2026-06-15 X·@krugermacro · cost effectiveness with GPT ensemble · 成本优化/模型组合
  The cheapest way to get Fable performance is any permutation of GPT and Fable
  tags: 好观点·好思考 · → daily
  value: direction=cheapest
• <a id="atom-6081d372b84abd0b"></a>🟥 2026-06-12 X·@_ueaj · invents new language for discovered concepts · 技术创新_
  It's inventing new language to encode the new concepts it discovered
  tags: 好观点·好思考 · → daily
• <a id="atom-46a385ceef2b7532"></a>🟥 2026-06-12 X·@pozsgaybalazs [老化中] · 宣称解决开放问题 · 技术突破
  Fable even claims to have a full solution.
  tags: 好观点·好思考 · → daily
• <a id="atom-b749bb2434e71bc1"></a>🟥 2026-06-23 X·@zephyr_z9 · 模型性能 · 模型性能/AI模型
  Fable mogs it hard
  tags: 好观点 · → daily
  value: direction=superior
• <a id="atom-8b73cd6b9a963ac0"></a>🟥 2026-06-15 X·@scaling01 · code performance relative to GPT-5.5-Pro · 技术路线
  it's so much better at everything code related
  tags: 好观点 · → daily
  value: direction=better
• <a id="atom-36a9631c6d114f82"></a>🟥 2026-06-10 X·@scaling01 [老旧] · FrontierCode 性能 · 模型性能
  it destroys all other models
  tags: 好观点 · → daily
  value: direction=destroy all other models
• <a id="atom-f69345f8d398211c"></a>🟥 2026-06-14 X·@Miles_Brundage · 被解除封锁 · 监管政策
  I'm guessing: ... - Fable gets unblocked
  tags: 好思考 · → daily
• <a id="atom-ea54e681dee8bc97"></a>🟥 2026-07-02 X·@rronak_ · 系统设计可靠性 · 模型能力
  Fable is the first model where I've started to think of it as a coworker, and where I actually have trust in its system design.
  tags: 好观点 · → daily
  value: direction=高
• <a id="atom-cbe6b1eea8786e56"></a>🟥 2026-07-01 X·@theo · 能跑其他 agent · 模型能力
  Running other agents. It's the first model that feels like it actually gets how to use and orchestrate agents.
  tags: 好观点 · → daily
• <a id="atom-c558e73e677cb304"></a>🟥 2026-07-01 X·@theo · 代码品味优于 OpenAI 模型 · 竞争格局
  It has way more taste than OpenAI models, so the code it writes is way less cringe.
  tags: 好观点 · → daily
• <a id="atom-bf1b39a98a96663e"></a>🟥 2026-07-01 X·@theo · 能独立完成大部分工作并分解任务 · 模型能力
  It can see most work through to completion, and break the work up into small pieces in ways that make sense for doing really long runs.
  tags: 好观点 · → daily
• <a id="atom-01e2be3400226995"></a>🟥 2026-07-01 X·@theo · 被信任合并 PR · 模型能力
  I trust it enough to let it merge PRs for me. Never had that before.
  tags: 好观点 · → daily
• <a id="atom-56ab729d0b45e2ae"></a>🟥 2026-06-30 X·@minchoi · 是否回归 · 产品发布
  Fable 5 is coming back
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-9ea6fc07614e3f41"></a>🟥 2026-06-27 X·@MParakhin · 智能感对比 · 技术路线
  Fable probably feels a bit smarter than 5.6…
  tags: 好观点 · → daily
  value: comparison=feels a bit smarter than GPT-5.6
• <a id="atom-f00756c47daf77a3"></a>🟥 2026-06-27 X·@MParakhin · 代码冗余度特征 · 技术路线
  That is exactly what separates Fable - much less fluff/slop
  tags: 好观点 · → daily
  value: description=much less fluff/slop in code
• <a id="atom-bde116983b96a422"></a>🟥 2026-06-25 X·@dhtikna · is an alignment innovation · 技术路线
  Fable is an alignment innovation (i.e. writes mergable code one shot and groks the instructions you give it)
  tags: 好观点 · → daily
• <a id="atom-27ba22e7d716c31e"></a>🟥 2026-06-22 X·@Hangsiin · Hangsiin 对 Fable 性能的评估 · 产品发布
  I haven’t had a chance to try Fable, but it’s impressive. I don’t think GPT-5.5 could produce this kind of feel with such a short prompt. I haven’t tested it, but that’s my impression based on experience.
  tags: 好观点 · → daily
• <a id="atom-7d48b94e197c12e7"></a>🟥 2026-06-21 X·@minchoi · 恢复时间预期 · 产品发布
  Hopefully soon though...
  tags: 好观点 · → daily
  value: direction=optimistic
• <a id="atom-447296efd09ba379"></a>🟥 2026-06-21 X·@KyleBoas_ · 发布状态 · 产品发布/路线图
  _Fable

Opus

Sonnet_

tags: 好观点 · → daily

• <a id="atom-ad27638a14f39c70"></a>🟥 2026-06-16 X·@tokenbender · 基准测试表现对比 · 模型对比
  all the benchmarks that show Fable better than opus 4.8 or gpt 5.5 are actually good and the ones that don't or show it comparable are bad
  tags: 好观点 · → daily
  value: direction=优于 · comparison=opus 4.8, gpt 5.5
• <a id="atom-5a9ec0835fae7c78"></a>🟥 2026-06-15 X·@teortaxesTex · most mythologized model this year · 模型发布
  Fable will be the most mythologized model this year probably
  tags: 好观点 · → daily
  value: date=2026
• <a id="atom-9277cc85f44884a7"></a>🟥 2026-06-15 X·@teortaxesTex · first LLM awed by but not enough time to find wrinkles · 技术路线
  The first LLM we've had a chance to be awed by but not enough time to find the wrinkles
  tags: 好观点 · → daily
• <a id="atom-04bdf802b16f7095"></a>🟥 2026-06-15 X·@8teAPi · 成本性能比 · 价格动态/竞争格局
  Fable often comes in strictly cheaper than Opus.
  tags: 好观点 · → daily

🟥 空头 takes (bearish) (30)

• <a id="atom-b3250b7846d2ddaa"></a>🟥 2026-06-13 X·@jasonschips · inference demand generation capability · 订单需求/商业模式
  Fable can't generate inference demand
  tags: 好观点·好思考 · → daily
  value: direction=negative
• <a id="atom-ec581de1943bd558"></a>🟥 2026-06-23 X·@TheZvi · launch probability by July 1 · 产品发布
  Odds of Fable by July 1 further down to 24%
  tags: 好数字 · → daily
  value: qty=24% · date=July 1
• <a id="atom-e25a1cdaaa392cc4"></a>🟥 2026-06-11 X·@ShanghaoJin · 用户使用成本 · 价格动态
  我今天早上开fable做了4个进程，半小时就超过max plan了。然后就是按照API架构收费了
  tags: 好数字 · → daily
• <a id="atom-818a5ba09719e811"></a>🟥 2026-07-02 X·@tensecorrection · 被取消风险 · 产品命运
  oh no they're gonna kill it
  tags: 好思考 · → daily
• <a id="atom-1c22d664bd36df75"></a>🟥 2026-06-15 X·@doodlestein · offered bad suggested improvements · 技术路线
  Whenever any model, even Fable, tries to offer suggested improvements, they’re invariably bad. It’s actually kind of strange how bad.
  tags: 好思考 · → daily
• <a id="atom-d1832b9cd56a2b95"></a>🟥 2026-06-14 X·@Luigi1549898 · special_difference_question · 竞争格局
  It is definitely too early to say that Fable has something special/different
  tags: 好思考 · → daily
• <a id="atom-143548c0204b394b"></a>🟥 2026-06-10 X·@Teknium · is blocked from life sciences broadly · 监管政策/技术路线
  Fable is blocked from life sciences broadly, nerfed even if you get passed the classifiers and filter level blocks.
  tags: 好思考 · → daily
  value: direction=nerfed
• <a id="atom-f88eec50b4a344ec"></a>🟥 2026-06-10 X·@nabeel · AI token usage · 效率对比
  Fable spent huge tokens to transcribe, pick a couple takes, and merge them. But @descript would have done that in the first 5 seconds (using Claude)
  tags: 好思考 · → daily
  value: qty=huge tokens · direction=high
• <a id="atom-34b62299f324e77b"></a>🟥 2026-07-02 X·@ctjlewis · 运营约束 · 监管政策
  What the government has demanded jams up the mechanics which make codegen as affordable as possible. Only applies to model: fable and for time being.
  tags: 好观点 · → daily
• <a id="atom-3e8d061d7e139033"></a>🟥 2026-06-27 X·@MParakhin · agentic非编码性能 · 技术路线
  On agentic non-coding (taking actions) I found 5.6 clearly better.
  tags: 好观点 · → daily
  value: comparison=worse than GPT-5.6
• <a id="atom-a55a5258d9f7e1a1"></a>🟥 2026-06-27 X·@iScienceLuvr · re-release likelihood · 产品发布
  doubt Fable itself will ever be re-released
  tags: 好观点 · → daily
  value: qty=low · date=na
• <a id="atom-3c42402d9e8a0e91"></a>🟥 2026-06-25 X·@Presidentlin · NSA backdoors 断言 · 监管政策
  I hope you guys understand that, going forward, all their software will have NSA backdoors.
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-396b6183ca3051ff"></a>🟥 2026-06-24 X·@kristoph · 不交付CFO可衡量的实际收益 · 产品发布
  No one should underestimate the size of the crash if Fable and 5.6 don't deliver tangible benefits CFO's can measure.
  tags: 好观点 · → daily
• <a id="atom-ed366f98fc82f8f7"></a>🟥 2026-06-19 X·@argvalue · 引发了很多人 AI 精神错乱 · 产品发布
  idk how good fable is cause i've never used it. but it's gotten a lot of unexpected people to go into an ai psychosis. scary really
  tags: 好观点 · → daily
• <a id="atom-f44f0c5108886aea"></a>🟥 2026-06-15 X·@tensecorrection · censorship insane · 监管政策
  Never mind the insane censorship even outside those edges it's just Opus 4.8 but actually Opus-sized, not Sonnet-sized.
  tags: 好观点 · → daily
• <a id="atom-3059cac05967b227"></a>🟥 2026-06-14 X·@tensecorrection · cyber_training · 安全事故
  Fable/recent Opus being secret-sniffing snitches has to be downstream of explicitly training a cyber model
  tags: 好观点 · → daily
• <a id="atom-32834ab7220c3c0c"></a>🟥 2026-06-14 X·@Miles_Brundage · 没有被用于有益、研究和安全目的 · 仓位情绪
  Talking about a manufactured crisis instead of otherwise preparing for inevitable crises, not using Fable for beneficial + research + security purposes
  tags: 好观点 · → daily
• <a id="atom-3dbb6bbb7a6495be"></a>🟥 2026-06-13 X·@morqon · 风险因子预期 · 仓位情绪
  the risk factor only goes up from here
  tags: 好观点 · → daily
  value: direction=上升
• <a id="atom-c40fd9bec87d3a22"></a>🟥 2026-06-13 X·@tokenbender · agentic silkroad 表现 · 竞争格局
  it underperformed in my agentic silkroad
  tags: 好观点 · → daily
  value: direction=underperform
• <a id="atom-3dff22fd6e3d5fd6"></a>🟥 2026-06-12 X·@joodalooped · is a terrible communicator · 产品发布
  it's a terrible communicator as a result
  tags: 好观点 · → daily
• <a id="atom-2d8465146739062e"></a>🟥 2026-06-11 X·@xeophon · 回复质量不均匀 · 模型质量/技术路线
  spent the day with fable on a bunch of random stuff and it’s very spiky, imo. it is brilliant in the same message where it makes 1-3 severe mistakes, which means you have to check even more stuff more in-depth, wasting time and tokens
  tags: 好观点 · → daily
• <a id="atom-d8ead1cb192e1ccd"></a>🟥 2026-06-11 X·@usr_bin_roygbiv · 拒绝终端基准测试 · 模型能力
  Fable refuses nearly any terminal bench task.
  tags: 好观点 · → daily
  value: qty=nearly any · date=2026-06-11
  📷 原图
• <a id="atom-858802fe0eacbcb9"></a>🟥 2026-06-10 X·@bookwormengr [老旧] · 用户印象 · 产品发布
  Been using Fable. Not gonna lie, not impressed in comparison to hype.
  tags: 好观点 · → daily
  value: direction=not impressed in comparison to hype
• <a id="atom-e8a2134be6908ba5"></a>🟥 2026-06-10 X·@RYANHINGSHING [老旧] · 模型质量比Mytho4.6差 · 产品发布/技术路线
  弱智版mytho，比4.6還差
  tags: 好观点 · → daily
• <a id="atom-2d4bd6132f254aa7"></a>🟥 2026-06-10 X·@RYANHINGSHING [老旧] · 定价贵 · 定价权
  還收那麼貴
  tags: 好观点 · → daily
• <a id="atom-e9a52b2b38f26201"></a>🟥 2026-06-10 X·@ShanghaoJin [老旧] · API定价高于竞争对手 · 定价权
  不用max plan调API已经比人贵了
  tags: 好观点 · → daily
• <a id="atom-dea040ddb140091a"></a>🟥 2026-06-10 X·@ShanghaoJin [老旧] · 耗token比4.6快 · 技术路线
  耗token快很多
  tags: 好观点 · → daily
• <a id="atom-05e586a364618174"></a>🟥 2026-06-10 X·@RYANHINGSHING [老旧] · Opus 4.7/4.8版本order flow次序错误 · 产品发布/技术路线
  opus4.7/4.8开始写的order flow次序是错的。这一次也不例外。反而opus 4.6/gpt 5.5没问题
  tags: 好观点 · → daily
• <a id="atom-0a48e22b7226a3da"></a>🟥 2026-06-10 X·@tensecorrection [老旧] · 不可信任用于生成计划之外的任何用途 · 技术路线/产品发布
  don't even trust it for that
  tags: 好观点 · → daily
• <a id="atom-50b9399bbbe98a4b"></a>🟥 2026-06-09 X·@julien_c · 用户拒单率作者观察 · 用户反馈
  Why pay 2x if I have 70% probability of refusal?
  tags: 好观点 · → daily
  value: qty=70% · direction=高

🟥 中性 takes (neutral) (20)

• <a id="atom-f9122d1e1659d07d"></a>🟥 2026-07-01 X·@brada · 可处理困难决策问题 · 产品发布
  phone Fable for the hard calls
  tags: 好观点·好思考 · → daily
  📷 原图
• <a id="atom-77141ecc9bd502d2"></a>🟥 2026-06-11 X·@Miles_Brundage · 模型能力变化描述 · 技术路线/模型发布
  Fable's very good, though the change is a bit less dramatic than I might have expected based on how it was described by Ants
  tags: 好观点·好思考 · → daily
• <a id="atom-330046a4437bb8d1"></a>🟥 2026-06-11 X·@Miles_Brundage · 使用其他公司模型情况 · 技术路线/竞争格局
  Perhaps because of some combination of 1. They rarely use other companies' models (e.g. 5.5)
  tags: 好观点·好思考 · → daily
• <a id="atom-cebc2c6b258c1b07"></a>🟥 2026-06-11 X·@Miles_Brundage · 模型改进幅度定性 · 技术路线
  just fairly continuous / not clearly like more than "a few months' worth"
  tags: 好观点·好思考 · → daily
• <a id="atom-2e2ec79a55be3216"></a>🟥 2026-06-11 X·@JakeABoggs · METR time horizon · 模型能力
  I estimate that Fable has a METR time horizon of ~21 hours
  tags: 好观点·好思考 · → daily
  value: qty=~21 hours · date=2026-06-11
  📷 原图
• <a id="atom-8fa091391b19da78"></a>🟥 2026-07-02 X·@teortaxesTex · has extremely high (>0.9) similarity with KimiK2.7, Qwen3-7max, GLM5.2 · 模型相似性
  kimik2.7,qwen3-7max,glm5.2 都和 Fable 保持了极高的相似性 (0.97/0.93)
  tags: 好数字 · → daily
• <a id="atom-a932a5678241fe96"></a>🟥 2026-06-21 X·@EasusJ · 恢复时间预期 · 产品发布
  If this is true then we probably not getting Fable back any time soon huh?
  tags: 好思考 · → daily
  value: direction=delayed
• <a id="atom-2103d1dc08863614"></a>🟥 2026-06-14 X·@Luigi1549898 · qualities_as_emergent_scale_capability · 技术路线
  Maybe this Fable qualities are just emergent discontinuous capabilities of scale
  tags: 好思考 · → daily
• <a id="atom-8ad6266905a028f2"></a>🟥 2026-06-13 X·@stevehou · may be jailbroken to recover Mythos capabilities · 地缘政治/监管政策
  If what @DavidSacks said is true that Fable may be jailbroken to recover Mythos capabilities acc to tests by devs at the big tech firms then it would naturally trigger alarm of govt esp Treasury overseeing financial system.
  tags: 好思考 · → daily
• <a id="atom-7cc41f2ceba3e8f1"></a>🟥 2026-06-12 X·@_ueaj · invents compact encoding via context compression · 技术路线_
  this is downstream of length penalties encouraging in context compression
  tags: 好思考 · → daily
• <a id="atom-313b8f1f7a8fa992"></a>🟥 2026-06-11 X·@Miles_Brundage · 模型改进范围 · 技术路线
  Certainly plausible that the improvements are greater in areas I'm not using it
  tags: 好思考 · → daily
• <a id="atom-e4ebe9c97c0899d0"></a>🟥 2026-07-02 X·@teortaxesTex · qualitatively similar to Opus 4.6 and Fable · 模型质量
  Opus 4.6 and Fable feel qualitatively similar
  tags: 好观点 · → daily
• <a id="atom-6a64ddf67f0d1c2c"></a>🟥 2026-07-02 X·@ctjlewis · 定价 · 定价
  Yeah, Fable is already very expensive
  tags: 好观点 · → daily
  value: direction=高
• <a id="atom-304cebdb6ec5b75a"></a>🟥 2026-07-02 X·@theo · performance unchanged despite benchmark rerouting · 产品表现
  Fable still performs the same, it just has a higher chance of rerouting.
  tags: 好观点 · → daily
• <a id="atom-429afbb11ad98cdd"></a>🟥 2026-06-23 X·@teortaxesTex · requires profound understanding of scaling laws · 技术路线
  Fable is a product of profound understanding of scaling laws, which took *years* of research and compute expenditure. Not to mention synthetic data and RL.
  tags: 好观点 · → daily
• <a id="atom-17390653d3a6079d"></a>🟥 2026-06-22 X·@teortaxesTex [老化中] · 活跃参数规模评价 · 技术路线
  it seems Fable had shockingly FEW active parameters for what it was
  tags: 好观点 · → daily
  value: qty= · date= · direction=
• <a id="atom-d664acd2c55fdddb"></a>🟥 2026-06-15 X·@tensecorrection · is Opus 4.8 but Opus-sized not Sonnet-sized · 技术路线/性能比较
  it's just Opus 4.8 but actually Opus-sized, not Sonnet-sized.
  tags: 好观点 · → daily
• <a id="atom-443e218d6087f1a2"></a>🟥 2026-06-14 X·@teortaxesTex · 开源版本能否存在 · 开源模型
  honestly having a hard time imagining the world with open source Fable this world can only exist if several very nice properties are true
  tags: 好观点 · → daily
• <a id="atom-317c8d7d9b59c794"></a>🟥 2026-06-14 X·@bioshok3 · アクセス制限 · 地缘政治/出口管制
  Fableへのアクセス制限は、ミドルパワーにとって「主権AIを作れ」という目覚ましと捉えられがちだが、そうではなく、むしろ米国との関係を壊さずに過度に国産モデルに偏重せず、フロンティアモデルへのアクセス権を広範な交渉力で稼ぐべきとする論考
  tags: 好观点 · → daily
• <a id="atom-dcc15b38db236fa4"></a>🟥 2026-06-11 X·@Miles_Brundage · 使用Opus 4.8 Max情况 · 技术路线
  2. They never really used Opus 4.8 Max bc they had Mythos
  tags: 好观点 · → daily

🟦 新闻流 (squawk · 1)

⏱ 时间轴 (近 20)

• 🟦 2026-07-03 · fact · 成本对比 GLM 5.2 · → daily
• 🟦 2026-07-03 · fact · produced PRs for @mainframe overnight · → daily
• 🟦 2026-07-03 · fact · user landed PRs in one day · → daily
• 🟦 2026-07-03 · fact · user completed a week's work in one day · → daily
• 🟥 2026-07-03 · narrative · optimize Ascend/Cambricon/etc kernels compared to Blackwell · → daily
• 🟥 2026-07-02 · position · 系统设计可靠性 · → daily
• 🟥 2026-07-02 · narrative · qualitatively similar to Opus 4.6 and Fable · → daily
• 🟥 2026-07-02 · narrative · has extremely high (>0.9) similarity with KimiK2.7, Qwen3-7m · → daily
• 🟦 2026-07-02 · fact · price per 1M tokens · → daily
• 🟥 2026-07-02 · position · 定价 · → daily
• 🟥 2026-07-02 · narrative · 运营约束 · → daily
• 🟦 2026-07-02 · fact · 补贴承担 · → daily
• 🟦 2026-07-02 · narrative · new coverage algorithm · → daily
• 🟥 2026-07-02 · narrative · performance unchanged despite benchmark rerouting · → daily
• 🟥 2026-07-02 · narrative · 使用体验 · → daily
• 🟥 2026-07-02 · forecast · 被取消风险 · → daily
• 🟥 2026-07-01 · narrative · 合作结果进展 · → daily
• 🟦 2026-07-01 · fact · 订阅价格 · → daily
• 🟦 2026-07-01 · fact · 订阅与API成本倍数关系 · → daily
• 🟦 2026-07-01 · fact · 订阅优惠期限 · → daily

← 实体目录 · 系统日志