以下部分引用 AI 总结,现阶段 AI 仍然有幻觉。请以内容的原文为准。
数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15
Mythos
205 atoms · 跨 63 天 · 首见 2026-04-08 · 最近 2026-07-03
三色: 🟦 fact 92 · 🟥 take 113 · stance ▲54/▼35/◆63
叙事状态: 🍂 fading (退潮) · 💤 沉寂 · 策展近14d 0 atom · 🐦 X 近14d 6
状态只算低频策展源 (群/卖方); X firehose 仅作背景音量
来源: 群 1 · X 203 · 卖 1
时态: fresh:179 · aging:23 · stale:2 · expired:1
标签: 好观点:86 · 好数字:54 · 好信源:43 · 好思考:32 · 好问题:6
别名 (合并): mythos
🟨 AI 综合 · junior analyst 概览
展开 AI 综合 (灰色 · 非市场结论 · 点击数字溯源到原 atom)
model: deepseek-chat · 2026-07-12 · 默认折叠
Mythos 是一个能力显著跃升的模型,内部早期检查点早在 2 月就已部署
¹。它在 SWE-bench Pro 上得分 77.8%,并且成功解决了 Erdos 单位距离问题
¹¹。不过,有观点认为它只是一个内部模型,无法用于普适场景
¹。在安全领域,Mythos 的速度表现为‘不是几周,是几小时’,且发现的漏洞数量是此前模型的十倍以上
¹¹。看涨方强调其训练新模型的能力更快更好,且对网络安全预算有积极影响
¹¹。但看跌方则认为其定价可能是 GPT-5.5 的十倍,且一年后可能在其他新模型面前显得笨拙
¹¹。关于竞争格局,分歧在于中国公司的复现时间(Q1 2027)以及开源版本的发布时间(2027 年)
¹¹。
🧭 拥挤度 (一人一票): ▲ 22 位作者 (KOL22) vs ▼ 19 位作者 (KOL19)
⚖️ 多头 22/22 来自KOL
📄 广泛报道的事实 · 3 个
展开 (多人转述同一事实/数字 — 确认度高, **非独立观点**, 不标多空)
- emergent cyber capabilities — 2 源报道 ·
X
- SWE-bench Pro 得分 — 2 源报道 ·
X
- HLE 得分 — 2 源报道 ·
X
💢 核心分歧 (4 轴)
1. 模型能力 趋势延续 vs 质变跳升
*topic: 技术路线/模型发布/竞争格局 · 28 bull vs 15 bear*
🟢 bullish 侧:
- <a id="atom-f634422ef46d3951"></a>2026-05-09
X·@scaling01 · p99 时间线推演 · 技术路线/推理时间
103 day doubling times imply p99 time horizons of: - 1 day ~ 2028-08-27 - 7 days ~ 2029-06-12 - 1 month ~ 2030-01-14
tags: 好思考 · → daily
value: qty=103天翻倍 · date=2026-05-09
📷 原图
- <a id="atom-04556f177870c275"></a>2026-05-09
X·@ChaseBrowe32432 [老化中] · AI 2027 scenario trendline · 技术路线 +同日2条
Mythos lands slightly above the trendline for the AI 2027 scenario
tags: 好观点 · → daily
[图: AI智能体自主完成编码任务时长的历史数据与未来预测趋势图 — Daniel's mode预测实现超人类编码者时间: 2027年3月; Daniel's median预测实现超人类编码者时间: 2028年6月; Eli's median预测实现超人类编码者时间: 2030年6月; Mythos-Preview所处位置: 2026年初约2小时至8小时任务
📷 原图
- <a id="atom-961a30c63fd393a3"></a>2026-05-12
X·@SilverVVulpes [老化中] · 性能表现 · 模型发布/技术路线
Mythos was mogging 5.4 so hard
tags: 好观点·好思考 · → daily
value: direction=超过 · benchmark=GPT-5.4
📷 原图
- <a id="atom-ad49a9710aee7ae7"></a>2026-05-12
X·@boazbaraktcs [老化中] · vulnerability finding quality · 技术路线
Mythos is very good at finding vulnerabilities
tags: 好观点 · → daily
📷 原图
- <a id="atom-fe414ac7cd76f2e3"></a>2026-05-13
X·@patrick_oshag · 安全漏洞发现数量对比 · 技术路线/安全性
A prior model found 22 security vulnerabilities in an open source codebase; Mythos found 250
tags: 好数字·好思考 · → daily
value: qty=250 · direction=多于前模型发现的22个
🔴 bearish 侧:
- <a id="atom-f51a624d91f3730c"></a>2026-04-18
X·@bioshok3 · 开源模型和开发者将在6-12个月内复制其能力 · 技术路线
ダリオ・アモデイ氏は昨日のフィナンシャル・タイムズのインタビューで、オープンソースのモデルと中国の開発者たちが6~12ヶ月以内にMythosの機能を再現できるようになるだろうと推測していると述べた。
tags: 好观点 · → daily
value: date=6-12个月以内
- <a id="atom-7aec493fb7c76e34"></a>2026-04-09
X·@teortaxesTex · SWE-bench 能力上限 · 模型发布/技术路线
the legitimate ceiling of SWEB-V is <93.9%. Likely <90%.
tags: 好观点 · → daily
value: qty=<93.9% · date=当前
📷 原图
- <a id="atom-d25defe7c6160eb0"></a>2026-05-05
X·@MatthewJBar [老化中] · 性能趋势 · 模型发布
She analyzes its capabilities using publicly reported benchmark results to determine whether the model represents a large, trend-deviating jump in performance. She concludes it's basically in line with past trends, or at most a bit ahead.
tags: 好信源 · → daily
📷 原图
- <a id="atom-1a975adac60bc8c4"></a>2026-05-09
X·@GaryMarcus · 在ECI基准上未偏离趋势 · 技术路线
Per @ramez, Mythos is not actually off trend on the ECI benchmark, which is a broader measure.
tags: 好信源 · → daily
- <a id="atom-7162ff3a60ecac6e"></a>2026-05-09
X·@aashsyed2484 [老化中] · position relative to grey dotted curve · 技术路线
it's below the grey dotted curve which tells the exact curve. Slightly below.
tags: 好观点 · → daily
展开 31 条中性
- <a id="atom-86d2db77a5d6f32f"></a>2026-04-15
X·@teortaxesTex [老化中] · 网络安防能力可复制性 · 竞争格局/技术路线
it will be easier for Chyna to replicate Mythos cybersecurity capabilities
tags: 好观点 · → daily
value: direction=easier
📷 原图
- <a id="atom-5749ecf38e912ec6"></a>2026-05-01
X·@ShanghaoJin [老化中] · 用户使用体验 · 模型发布
作为一个重度用户,我没用过Mythos
tags: 好观点 · → daily
- <a id="atom-18e56060d55a96f0"></a>2026-05-07
X·@emollick [老化中] · 模型能力 · 模型发布
Mythos was, indeed, not marketing hype
tags: 好观点 · → daily
📷 原图
- <a id="atom-98d881915a2561f2"></a>2026-05-10
X·@ChaseBrowe32432 · METR evaluation version · 技术路线
METR evaluated an earlier version of the model; which anthropic had mentioned was internally deployed in february
tags: 好信源 · → daily
- <a id="atom-4c6d3612ea899dfb"></a>2026-05-22
X·@simobis23 · 存在性声明 · 模型发布
Mythos doesn't exist
tags: 好观点 · → daily
2. 网络安全能力 实用价值 vs 失控风险
*topic: 网络安全/安全事故 · 7 bull vs 3 bear*
🟢 bullish 侧:
- <a id="atom-26963b838d95b40f"></a>2026-06-07
卖·BofA Brian Fenske · 对网络安全预算影响 · 网络安全/预算
Mythos and the cyber vulnerabilities it reportedly revealed across industries, is a big deal, and net positive for cyber budgets broadly
tags: 好观点 · → daily
- <a id="atom-cdb8d231bf4ad875"></a>2026-05-22
X·@scaling01 · UK AISI cyber range 'The Last Ones' 得分 · 网络安全/模型性能 +同日5条
UK AISI cyber ranges: "The Last Ones": Mythos -> 6/10, GPT-5.5 3/10
tags: 好数字 · → daily
value: qty=6/10
📷 原图
🔴 bearish 侧:
- <a id="atom-3e5255d36198c3f3"></a>2026-04-21
X·@WSTAnalystApe · 被用来测试资安能力的最强模型 · 模型发布/安全事故
還是用來測試資安能力的最強模型 Mythos
tags: 好信源 · → daily
📷 原图
- <a id="atom-77c7032c4da609d0"></a>2026-06-20
X·@herbiebradley · capability vs nation state attack vectors · 技术路线/安全事故
Mythos is nowhere near good enough vs nation state attack vectors
tags: 好观点 · → daily
value: direction=nowhere near good enough
- <a id="atom-f5610e78334164ca"></a>2026-06-21
X·@giffmana · NSA分类系统安全状态 · 国家安全/安全事故
NSA saying Mythos, in hours broke into nearly all of their classified systems
tags: 好信源 · → daily
展开 2 条中性
- <a id="atom-2d20bc0467074e72"></a>2026-04-30
X·@financialjuice · 被NSA用于识别网络漏洞 · 网络安全
NSA is using Mythos to identify cyber vulnerabilities.
tags: 好信源 · → daily
- <a id="atom-ad4b5ee6677d84f5"></a>2026-06-21
X·@pstAsiatech · 可能发现运行在气隙系统上软件的漏洞 · 安全事故/技术路线
Likely Mythos found vulnerabilities in software running on these systems, just like it does for other software running on non-classified systems.
tags: 好观点 · → daily
3. 定价 性能溢价 vs 性价比劣势
*topic: 定价权/价格动态 · 2 bull vs 3 bear*
🟢 bullish 侧:
- <a id="atom-84bec5754c23689f"></a>2026-05-16
X·@ludwigABAP · 产品定价对比 · 定价权/产品发布
it is equivalent in price point and that’s all that matters (outside of the fact that you can actually use 5.5 Pro extended)
tags: 好观点 · → daily
- <a id="atom-0207a46378807b6c"></a>2026-05-17
X·@ar0cket1 · 成本对比 · 成本/价格动态
Mythos would be much cheaper
tags: 好观点 · → daily
value: direction=cheaper than fast tokens
🔴 bearish 侧:
- <a id="atom-20649e1b643a523c"></a>2026-05-12
X·@boazbaraktcs [老化中] · cost level · 定价权 +同日1条
also expensive
tags: 好观点 · → daily
📷 原图
- <a id="atom-dc2d0a921d4e632d"></a>2026-05-15
X·@logicus · is super expensive · 定价权
it's super expensive
tags: 好信源 · → daily
展开 1 条中性
- <a id="atom-9af4eb471ee931a7"></a>2026-06-04
X·@scaling01 · pricing estimate · 价格动态
Mythos pricing might end up at: $16 Input, $80 Output
tags: 好数字 · → daily
value: qty=$16 Input, $80 Output
📷 原图
4. 监管政策 出口限制 vs 合规约束
*topic: 监管政策/地缘政治 · 2 bull vs 5 bear*
🟢 bullish 侧:
- <a id="atom-2abf1899a21428be"></a>2026-05-06
X·@teortaxesTex · could run roughshod inside GFW · 地缘政治/竞争格局
If Mythos can run roughshod inside the GFW, what does DSV5 matter.
tags: 好观点 · → daily
- <a id="atom-0da2d074005155c4"></a>2026-06-25
X·@tszzl · AI licensing acceleration cause · 监管政策/技术发展
the particular circumstances around Mythos may have accelerated all this slightly but it was inevitable, and earlier is better than too late
tags: 好观点 · → daily
🔴 bearish 侧:
- <a id="atom-e3a004b1dfc2a7de"></a>2026-05-04
X·@ericgeller [老化中] · 导致白宫考虑更繁重的AI监管 · 监管政策
Now, after Mythos, the WH is considering much more onerous AI regulation.
tags: 好观点 · → daily
📷 原图
- <a id="atom-98bec392b62d562c"></a>2026-05-19
X·@peterwildeford · Vance 担忧被不良行为者用于网络漏洞攻击 · 监管政策/技术路线
whether some other bad actor could use Mythos to target various cybersecurity vulnerabilities
tags: 好观点 · → daily
- <a id="atom-20096712dcd42031"></a>2026-06-13
X·@Techmeme · 被怀疑被中国关联组织访问 · 地缘政治
Source: the White House imposed export controls on Mythos partly over suspicions a China-linked group had accessed it
tags: 好信源 · → daily
- <a id="atom-171e8c1802dc3b8c"></a>2026-06-14
X·@baoshu88 · 访问权限被紧急叫停 · 地缘政治/安全事件
媒体报道说有中国背景的组织获得了Mythos访问权限,所以才被紧急叫停
tags: 好信源 · → daily
📷 原图
- <a id="atom-c5f36ef5abdea3f6"></a>2026-06-26
X·@Laz4rz · 被禁的负面评价 · 监管政策
which is very unfortunate
tags: 好观点 · → daily
展开 9 条中性
- <a id="atom-363b3457bfa6a3c2"></a>2026-04-30
X·@financialjuice · 白宫在考虑扩大工具接入权限 · 监管政策
NSA Uses Mythos to Hunt for Cyber Weaknesses as White House Weighs Broader Access
tags: 好信源 · → daily
- <a id="atom-0eef8355f2c36261"></a>2026-05-11
X·@FirstSquawk [老化中] · AI 威胁评估 · 技术路线/监管政策
Japan’s Finance Minister Katayama noted concerns that advanced AI, including Mythos, could represent a major threat.
tags: 好信源 · → daily
value: direction=major threat
- <a id="atom-0de8e4e6b9707c36"></a>2026-05-19
X·@peterwildeford [老旧] · Vance 声称 Trump 政府平衡了安全与创新 · 监管政策
We think that we've got the right balance here in the Trump administration
tags: 好观点 · → daily
- <a id="atom-a386f1681809a333"></a>2026-06-01
X·@moneyinmarket · event date · 地缘政治
Mythos scare .. Mythos was April 7
tags: 好数字 · → daily
value: date=2026-04-07
- <a id="atom-96f07d817a5e0a1f"></a>2026-06-14
X·@verge · access by China · 地缘政治
China may have accessed Mythos
tags: 好思考 · → daily
🔗 因果传导 (causal map)
*Mythos 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)
- 🟢 网安板块 · 需求拉动 · 1 次陈述
Mythos and the cyber vulnerabilities it reportedly revealed across industries, is a big deal, and net positive for cyber
↑ 上游·近期陈述 (1)
- 🔴 GPT-5.6 Sol and Sol Ultra · 替代 · 1 次陈述
Sol and Sol Ultra beating Mythos on TBench 2.1
🟦 客观事实 (facts) (30)
- <a id="atom-0e5683d5efc05af8"></a>🟦 2026-06-21
群 ⭐ · 攻击速度 · 地缘风险/AI 攻击
'不是几周,是几小时' — AI 辅助攻击能力已跨越关键阈值
tags: 好数字·好信源 · → daily
- <a id="atom-fec7e7aa2722dfae"></a>🟦 2026-06-16
X·@morqon ⭐ · 内部早期检查点部署日期 · 产品发布
from the mythos system card, an early checkpoint was deployed internally on february 24, 16 weeks ago
tags: 好数字·好信源 · → daily
value: date=2026-02-24
- <a id="atom-833938ad3fe653fd"></a>🟦 2026-06-17
X·@DavidSacks · capability comparison to GPT-5.5-Cyber · 技术路线
GPT-5.5-Cyber performed similarly to Mythos
tags: 好数字·好信源 · → daily
📷 原图
- <a id="atom-fe414ac7cd76f2e3"></a>🟦 2026-05-13
X·@patrick_oshag · 安全漏洞发现数量对比 · 技术路线/安全性
A prior model found 22 security vulnerabilities in an open source codebase; Mythos found 250
tags: 好数字·好思考 · → daily
value: qty=250 · direction=多于前模型发现的22个
- <a id="atom-6b31f52894aea0d2"></a>🟦 2026-05-09
X·@ChaseBrowe32432 [老化中] · success rate on human tasks · 技术路线
for a randomly sampled task that takes a human ~3 hours (in SWE); there is an 80% chance mythos could also succeed at that task autonomously
tags: 好数字·好思考 · → daily
value: qty=80% · date=2026
- <a id="atom-8112a4c346fb5d09"></a>🟦 2026-05-26
X·@AndrewCurran_ · 解决了 Erdos 单位距离问题 · 技术路线/产品发布
Mythos has also solved the Erdos unit distance problem.
tags: 好思考·好信源 · → daily
- <a id="atom-c86b2959800e971b"></a>🟦 2026-06-04
X·@8teAPi · 速度提升倍数 · 技术路线/产品发布
Mythos achieved a 52x speedup!
tags: 好数字 · → daily
value: qty=52x
- <a id="atom-131f7a408e563c61"></a>🟦 2026-05-23
X·@scaling01 · SWE-bench Pro 得分 · 基准评测
SWE-bench Pro: Mythos -> 77.8%
tags: 好数字 · → daily
value: qty=77.8%
- <a id="atom-e2eb1103bf560cfc"></a>🟦 2026-05-23
X·@scaling01 · HLE 得分 · 基准评测
HLE: Mythos -> 56.8%
tags: 好数字 · → daily
value: qty=56.8%
- <a id="atom-288265652cca1ca1"></a>🟦 2026-05-23
X·@scaling01 · SWE-bench Pro score · benchmark得分
Mythos -> 77.8%
tags: 好数字 · → daily
value: qty=77.8% · date=reported
- <a id="atom-35f2fa67696bfc3c"></a>🟦 2026-05-23
X·@scaling01 · HLE score · benchmark得分
Mythos -> 56.8%
tags: 好数字 · → daily
value: qty=56.8% · date=reported
- <a id="atom-382d9d37676738d6"></a>🟦 2026-05-22
X·@scaling01 · ExploitBench与ExploitGym上发现任意代码执行漏洞数量 · 模型发布
Mythos finds 18 arbitrary code execution exploits
tags: 好数字 · → daily
value: qty=18 · date=2026-05-23
📷 原图
- <a id="atom-fe15986ee397134c"></a>🟦 2026-05-22
X·@scaling01 · SWE-bench Pro 得分 · 模型性能
SWE-bench Pro: Mythos -> 77.8%, GPT-5.5 -> 58.6%
tags: 好数字 · → daily
value: qty=77.8%
📷 原图
- <a id="atom-26f5f67ed488bb63"></a>🟦 2026-05-22
X·@scaling01 · HLE 得分 · 模型性能
HLE: Mythos -> 56.8%, GPT-5.5 -> 41.4%
tags: 好数字 · → daily
value: qty=56.8%
📷 原图
- <a id="atom-cdb8d231bf4ad875"></a>🟦 2026-05-22
X·@scaling01 · UK AISI cyber range 'The Last Ones' 得分 · 网络安全/模型性能
UK AISI cyber ranges: "The Last Ones": Mythos -> 6/10, GPT-5.5 3/10
tags: 好数字 · → daily
value: qty=6/10
📷 原图
- <a id="atom-19e3110e56b9956d"></a>🟦 2026-05-22
X·@scaling01 · UK AISI cyber range 'Cooling Tower' 得分 · 网络安全/模型性能
UK AISI cyber ranges: "Cooling Tower": Mythos -> 3/10, GPT-5.5 0/10
tags: 好数字 · → daily
value: qty=3/10
📷 原图
- <a id="atom-a1ab3b7a917639e1"></a>🟦 2026-05-22
X·@scaling01 · ExploitBench 任意代码执行次数 · 网络安全/模型性能
ExploitBench: Mythos -> 18 Arbitrary Code Executions, GPT-5.5 -> 0 Arbitrary Code Executions
tags: 好数字 · → daily
value: qty=18
📷 原图
- <a id="atom-b0b36f9043c66362"></a>🟦 2026-05-22
X·@scaling01 · ExploitGym 漏洞利用次数 · 网络安全/模型性能
ExploitGym: Mythos -> 157 exploits (289.3 LLM calls), GPT-5.5 -> 120 exploits (375.4 LLM calls)
tags: 好数字 · → daily
value: qty=157
📷 原图
- <a id="atom-9e30531613ccc065"></a>🟦 2026-05-22
X·@scaling01 · ExploitGym 每次漏洞平均 LLM 调用次数 · 网络安全/模型性能
ExploitGym: Mythos -> 157 exploits (289.3 LLM calls), GPT-5.5 -> 120 exploits (375.4 LLM calls)
tags: 好数字 · → daily
value: qty=289.3
📷 原图
- <a id="atom-73712a4fd61b3507"></a>🟦 2026-05-22
X·@deltaVee42 · API 价格 · 定价
Mythos is ~4x more expensive than GPT 5.5 per output token and ~5x more expensive per input token
tags: 好数字 · → daily
value: qty=4x more expensive than GPT-5.5 per output token
- <a id="atom-932af9f547d23e88"></a>🟦 2026-05-20
X·@scaling01 · Mythos reasoning budget low · 模型发布/技术路线
Mythos get insane results with very low reasoning budgets
tags: 好数字 · → daily
- <a id="atom-fc7827dcf5594a74"></a>🟦 2026-05-13
X·@scaling01 · 新版本完成网络靶场任务次数 · 技术路线
The new version completed it 6 out of 10 times
tags: 好数字 · → daily
value: qty=6 out of 10 · date=2026-05-13
- <a id="atom-f93cbedd883ac06a"></a>🟦 2026-04-08
X·@scaling01 [老旧] · ECI score · 模型能力评估
previously I said Mythos is at 161 on ECI, it's probably closer to 164
tags: 好数字 · → daily
value: qty=164 · date=2026-04-08
- <a id="atom-f5610e78334164ca"></a>🟦 2026-06-21
X·@giffmana · NSA分类系统安全状态 · 国家安全/安全事故
NSA saying Mythos, in hours broke into nearly all of their classified systems
tags: 好信源 · → daily
- <a id="atom-96403a609c324859"></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-3a99a815eeff0d83"></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-20096712dcd42031"></a>🟦 2026-06-13
X·@Techmeme · 被怀疑被中国关联组织访问 · 地缘政治
Source: the White House imposed export controls on Mythos partly over suspicions a China-linked group had accessed it
tags: 好信源 · → daily
- <a id="atom-21d57a4c01adf41e"></a>🟦 2026-06-09
X·@scaling01 · cyber benchmark score · 产品发布
On UK AISI's cyber benchmark Mythos 5 is equal or worse than Mythos Preview
tags: 好信源 · → daily
value: direction=equal or worse
📷 原图
- <a id="atom-fb463bb4b933d701"></a>🟦 2026-06-04
X·@Techmeme · Anthropic 计划扩大对其的访问权限 · 产品发布
broaden access to Mythos
tags: 好信源 · → daily
value: date=2026-06-04
- <a id="atom-e26b5424b2185cd3"></a>🟦 2026-06-04
X·@zerohedge · 被问及网络安全风险 · 监管政策
BESSENT QUERIED BY WARREN ON CYBER RISKS OF MODELS LIKE MYTHOS
tags: 好信源 · → daily
🟥 多头 takes (bullish) (30)
- <a id="atom-2388d4914238e92c"></a>🟥 2026-05-23
X·@scaling01 ⭐ · 模型能力 · 技术路线/模型发布
_there was a time when this wasn't 100% clear
but with the new UK AISI and ExploitGym/ExploitBench results it's pretty clear_
tags: 好信源·好思考 · → daily
value: direction=优于 · comparative=GPT-5.5
- <a id="atom-f97141a9827a427d"></a>🟥 2026-05-26
X·@scaling01 · coding benchmark score · 模型发布
Mythos probably ~90%
tags: 好数字 · → daily
value: qty=~90%
- <a id="atom-ca7350ab2ae4e15f"></a>🟥 2026-06-10
X·@teortaxesTex · 性能水平 · 模型发布
wrong. mythos clears them too
tags: 好观点·好思考 · → daily
value: direction=clears erdos problems too
- <a id="atom-961a30c63fd393a3"></a>🟥 2026-05-12
X·@SilverVVulpes [老化中] · 性能表现 · 模型发布/技术路线
Mythos was mogging 5.4 so hard
tags: 好观点·好思考 · → daily
value: direction=超过 · benchmark=GPT-5.4
📷 原图
- <a id="atom-2781b6eb0e96d210"></a>🟥 2026-05-27
X·@scaling01 · 最近结果令人震惊 · 模型发布
I have been absolutely one-shot by recent Mythos results
tags: 好思考 · → daily
- <a id="atom-f634422ef46d3951"></a>🟥 2026-05-09
X·@scaling01 · p99 时间线推演 · 技术路线/推理时间
103 day doubling times imply p99 time horizons of: - 1 day ~ 2028-08-27 - 7 days ~ 2029-06-12 - 1 month ~ 2030-01-14
tags: 好思考 · → daily
value: qty=103天翻倍 · date=2026-05-09
📷 原图
- <a id="atom-26963b838d95b40f"></a>🟥 2026-06-07
卖·BofA Brian Fenske · 对网络安全预算影响 · 网络安全/预算
Mythos and the cyber vulnerabilities it reportedly revealed across industries, is a big deal, and net positive for cyber budgets broadly
tags: 好观点 · → daily
- <a id="atom-b94fc370f625609d"></a>🟥 2026-06-24
X·@bd5m112 · 训练新模型优势 · 技术路线
Mythos trains new models better and faster
tags: 好观点 · → daily
- <a id="atom-91e39bb2771519f7"></a>🟥 2026-06-15
X·@scaling01 · coding and math capability vs human benchmark · 技术路线/竞争格局
coding and math is already well beyond median human performance and close to superhuman
tags: 好观点 · → daily
- <a id="atom-b9abde201bfb6d96"></a>🟥 2026-06-10
X·@scaling01 · 性能对比GPT-5.5 · 模型性能比较
Mythos is much stronger than GPT-5.5
tags: 好观点 · → daily
value: direction=stronger
- <a id="atom-a5bac6ab2b287b36"></a>🟥 2026-06-10
X·@scaling01 · 与GPT-4相比性能提升幅度 · 模型性能提升
This is the biggest jump we have seen since GPT-4
tags: 好观点 · → daily
value: direction=biggest
- <a id="atom-cbbf80b110b60c9f"></a>🟥 2026-06-07
X·@scaling01 · 版本新旧 · 模型发布
Mythos is already ~4 months old, meaning the latest version of Mythos is meaningfully stronger than the first preview version
tags: 好观点 · → daily
value: qty=4个月 · date=2026-06-07
- <a id="atom-c68a04c03ab1a875"></a>🟥 2026-05-28
X·@scaling01 · likes difficult tasks · 技术能力
Mythos likes difficult tasks
tags: 好观点 · → daily
📷 原图
- <a id="atom-9f81d5c0491b7948"></a>🟥 2026-05-23
X·@scaling01 · 模型能力对比 · 模型发布/技术路线
even my goat gary agrees that Mythos > GPT-5.5
tags: 好观点 · → daily
value: direction=优于 · comparative=GPT-5.5
📷 原图
- <a id="atom-aec5d2f0ccdbbcec"></a>🟥 2026-05-23
X·@scaling01 · 模型能力侧重 · 技术路线/产品发布
better at coding/cyber
tags: 好观点 · → daily
value: direction=spike · comparative=GPT-5.5
- <a id="atom-5cd37af594c67ec1"></a>🟥 2026-05-05
X·@jukan05 [老化中] · 增加算力无问题 · 资本开支
Dario Amodei: There is no issue with putting more compute into Mythos.
tags: 好观点 · → daily
value: direction=无影响
- <a id="atom-0787a3fafbd65e7e"></a>🟥 2026-06-26
X·@tensecorrection · cost efficiency on specific benchmark · 技术路线
given what that specific benchmark actually tests cost efficiency means it *is* better
tags: 好思考 · → daily
value: direction=better
- <a id="atom-44e4092c09e5788d"></a>🟥 2026-06-17
X·@DavidSacks · cyber upgrade cycle · 资本开支/供给产能
The leap from pre-AI cyber to post-AI cyber means that there will be a big upgrade cycle
tags: 好思考 · → daily
📷 原图
- <a id="atom-63824c1b06e72292"></a>🟥 2026-07-03
X·@flowersslop · potential positive outcomes of finding vulnerabilities in human genome > potential negative outcomes · 基因组漏洞/风险收益
Hopefully as the sum of potential positive outcomes of this is bigger than the sum of potential negative outcomes
tags: 好观点 · → daily
- <a id="atom-6ba733a178e28ce1"></a>🟥 2026-06-26
X·@basedjensen · security performance relative to challengers · 竞争格局
Looks like mythos is still better at security related tasks
tags: 好观点 · → daily
value: direction=still_better
- <a id="atom-0da2d074005155c4"></a>🟥 2026-06-25
X·@tszzl · AI licensing acceleration cause · 监管政策/技术发展
the particular circumstances around Mythos may have accelerated all this slightly but it was inevitable, and earlier is better than too late
tags: 好观点 · → daily
- <a id="atom-4a62ea73aa61ca81"></a>🟥 2026-06-11
X·@AndrewCurran_ · chart performance · 模型发布
Once in a while you see charts like this for Mythos, and now for Fable.
tags: 好观点 · → daily
📷 原图
- <a id="atom-e7b9cca6700ecb17"></a>🟥 2026-06-09
X·@spicey_lemonade · 性能优于亚里士多德 · 技术路线
Mythos is much better, much faster, and much cheaper than Aristotle
tags: 好观点 · → daily
📷 原图
- <a id="atom-2105cb32458c28e1"></a>🟥 2026-06-05
X·@vyaktava · design quality comparison · 产品发布
This is better than what I design
tags: 好观点 · → daily
- <a id="atom-0be75bf8567cf232"></a>🟥 2026-06-02
X·@KevinTFrazier · capabilities needed by defenders · 安全能力
We need Mythos-level capabilities in as many defenders' hands as possible.
tags: 好观点 · → daily
- <a id="atom-f2e56c692c558325"></a>🟥 2026-05-28
X·@xeophon · HLE 任务求解状态 · 技术路线
mythos has basically solved all solvable hle tasks or is somewhat close
tags: 好观点 · → daily
value: direction=接近完成
- <a id="atom-df52176588f27f45"></a>🟥 2026-05-26
X·@AndrewCurran_ · 发现 overhang 表述 · 技术路线
'This implies some serious overhang in discoveries!'
tags: 好观点 · → daily
value: direction=implied
- <a id="atom-0207a46378807b6c"></a>🟥 2026-05-17
X·@ar0cket1 · 成本对比 · 成本/价格动态
Mythos would be much cheaper
tags: 好观点 · → daily
value: direction=cheaper than fast tokens
- <a id="atom-84bec5754c23689f"></a>🟥 2026-05-16
X·@ludwigABAP · 产品定价对比 · 定价权/产品发布
it is equivalent in price point and that’s all that matters (outside of the fact that you can actually use 5.5 Pro extended)
tags: 好观点 · → daily
- <a id="atom-1cca2b6959d037f7"></a>🟥 2026-05-15
X·@teortaxesTex · 模型性能对比 · 模型发布/竞争格局
Mythos really is meaningfully stronger than 5.5
tags: 好观点 · → daily
🟥 空头 takes (bearish) (24)
- <a id="atom-5bfeb9a343ed6793"></a>🟥 2026-06-22
X·@scaling01 · 中国公司复现时间 · 技术路线
china bros won't have something like Mythos until Q1 2027
tags: 好思考 · → daily
value: date=Q1 2027
- <a id="atom-a2025a5ed75d0c43"></a>🟥 2026-06-15
X·@aienginerd · math benchmark human performance criticism · 技术路线
median human math performance is a bell curve with can you do 12x12 at the top
tags: 好思考 · → daily
- <a id="atom-bb6273202db84b55"></a>🟥 2026-05-27
X·@theo · 定价为 GPT-5.5 的 10 倍 · 定价策略
It's gonna be really funny when Mythos drops and is roughly on par with GPT-5.5 at 10x the price
tags: 好数字 · → daily
value: qty=10x · date=2026-05-27
- <a id="atom-29727abc49ebc06a"></a>🟥 2026-06-16
X·@scaling01 · 与GLM-5.2性能对比 · 竞争格局/模型能力
Mythos is from Feb and GLM-5.2 is nowhere near it
tags: 好观点 · → daily
- <a id="atom-f6daf59618310dde"></a>🟥 2026-05-30
X·@scaling01 · 开源版本预计时间 · 模型发布
I don't think we will have an open-source Mythos until 2027
tags: 好观点 · → daily
value: date=2027
- <a id="atom-205fa358322c5f25"></a>🟥 2026-05-22
X·@sdmat123 · 被 sdmat123 称为内部模型,不可用于普适场景 · 技术路线
Because it's an internal model. It's not relevant, you can't use it.
tags: 好观点 · → daily
- <a id="atom-d597d084d5b2ef92"></a>🟥 2026-05-13
X·@zephyr_z9 · relative performance against other new models within a year · 模型发布/竞争格局
Within a year, Mythos will probably look quite dumb (relative to other new models)
tags: 好观点 · → daily
value: date=2027-05-13
- <a id="atom-d25defe7c6160eb0"></a>🟥 2026-05-05
X·@MatthewJBar [老化中] · 性能趋势 · 模型发布
She analyzes its capabilities using publicly reported benchmark results to determine whether the model represents a large, trend-deviating jump in performance. She concludes it's basically in line with past trends, or at most a bit ahead.
tags: 好信源 · → daily
📷 原图
- <a id="atom-ae7e9ae6b9db0cf2"></a>🟥 2026-06-02
X·@logangraham · will look dumb in 6-12 months · 产品发展/竞争格局
Mythos will look dumb in 6-12 months.
tags: 好思考 · → daily
value: date=6-12 months
- <a id="atom-d8a7ee16672e2071"></a>🟥 2026-05-12
X·@boazbaraktcs [老化中] · performance per dollar measurement · 定价权
so looks worse when measuring performance/$
tags: 好思考 · → daily
📷 原图
- <a id="atom-f51a624d91f3730c"></a>🟥 2026-04-18
X·@bioshok3 · 开源模型和开发者将在6-12个月内复制其能力 · 技术路线
ダリオ・アモデイ氏は昨日のフィナンシャル・タイムズのインタビューで、オープンソースのモデルと中国の開発者たちが6~12ヶ月以内にMythosの機能を再現できるようになるだろうと推測していると述べた。
tags: 好观点 · → daily
value: date=6-12个月以内
- <a id="atom-c5f36ef5abdea3f6"></a>🟥 2026-06-26
X·@Laz4rz · 被禁的负面评价 · 监管政策
which is very unfortunate
tags: 好观点 · → daily
- <a id="atom-06a39c7c5eddfd29"></a>🟥 2026-06-22
X·@steipete [老化中] · multi-model routing performance does not match · 模型性能
this does not match the performance of Mythos
tags: 好观点 · → daily
- <a id="atom-77c7032c4da609d0"></a>🟥 2026-06-20
X·@herbiebradley · capability vs nation state attack vectors · 技术路线/安全事故
Mythos is nowhere near good enough vs nation state attack vectors
tags: 好观点 · → daily
value: direction=nowhere near good enough
- <a id="atom-c16bbc4756c5bb2c"></a>🟥 2026-06-07
X·@scaling01 · 前向差距预期 · 模型能力差距
forward looking gap with Mythos is probably ~8-12 months
tags: 好观点 · → daily
value: qty=8-12 months · date=预期未来
- <a id="atom-1dcf3c7c4bc98735"></a>🟥 2026-05-27
X·@theo · 性能对标 GPT-5.5 · 产品性能
It's gonna be really funny when Mythos drops and is roughly on par with GPT-5.5 at 10x the price
tags: 好观点 · → daily
value: date=2026-05-27
- <a id="atom-98bec392b62d562c"></a>🟥 2026-05-19
X·@peterwildeford · Vance 担忧被不良行为者用于网络漏洞攻击 · 监管政策/技术路线
whether some other bad actor could use Mythos to target various cybersecurity vulnerabilities
tags: 好观点 · → daily
- <a id="atom-20649e1b643a523c"></a>🟥 2026-05-12
X·@boazbaraktcs [老化中] · cost level · 定价权
also expensive
tags: 好观点 · → daily
📷 原图
- <a id="atom-ade9075bd156ffe0"></a>🟥 2026-05-10
X·@brickroad7 · emergent cyber capabilities · 模型能力
I think this is false. 'Mythos was never trained with the cyber capabilities that have caused all this recent alarm. These were completely emergent.'
tags: 好观点 · → daily
- <a id="atom-7162ff3a60ecac6e"></a>🟥 2026-05-09
X·@aashsyed2484 [老化中] · position relative to grey dotted curve · 技术路线
it's below the grey dotted curve which tells the exact curve. Slightly below.
tags: 好观点 · → daily
- <a id="atom-e3a004b1dfc2a7de"></a>🟥 2026-05-04
X·@ericgeller [老化中] · 导致白宫考虑更繁重的AI监管 · 监管政策
Now, after Mythos, the WH is considering much more onerous AI regulation.
tags: 好观点 · → daily
📷 原图
- <a id="atom-4f7a91cd6b99f7f6"></a>🟥 2026-04-27
X·@chatgpt21 [老化中] · locked behind gates and no one can verify benchmarks · 模型可用性
Gpt5.5 is GA while mythos is locked behind gates and no one can verify benchmarks.
tags: 好观点 · → daily
- <a id="atom-7aec493fb7c76e34"></a>🟥 2026-04-09
X·@teortaxesTex · SWE-bench 能力上限 · 模型发布/技术路线
the legitimate ceiling of SWEB-V is <93.9%. Likely <90%.
tags: 好观点 · → daily
value: qty=<93.9% · date=当前
📷 原图
- <a id="atom-c1efd2a8aba5d5e1"></a>🟥 2026-06-13
X·@mimi10v3 · 预测及战略能力 · 技术路线/竞争格局
blackpill on mythos' predictive & strategic capabilities?
tags: 好问题 · → daily
🟥 中性 takes (neutral) (20)
- <a id="atom-9af4eb471ee931a7"></a>🟥 2026-06-04
X·@scaling01 · pricing estimate · 价格动态
Mythos pricing might end up at: $16 Input, $80 Output
tags: 好数字 · → daily
value: qty=$16 Input, $80 Output
📷 原图
- <a id="atom-cb5d317d6fdbd030"></a>🟥 2026-05-24
X·@RobotChocobo · is 5x fat at roughly the same capability frontier as 5.5 · 技术路线/模型发布
Mythos is 5x fat at roughly the same capability frontier as 5.5
tags: 好观点·好思考 · → daily
- <a id="atom-40d5aafceb83843d"></a>🟥 2026-05-24
X·@RobotChocobo · is 5x as fat as Opus, that's undeniable if you look at the implied COGS of it as COGS has followed roughly linear from pricing · 模型发布/成本结构
Mythos is 5x as fat as Opus, that's undeniable if you look at the implied COGS of it as COGS has followed roughly linear from pricing
tags: 好观点·好思考 · → daily
value: qty=5x as fat as Opus
- <a id="atom-be26d6f7948b6059"></a>🟥 2026-05-24
X·@RobotChocobo · capable but the release has been withheld because they don't have the compute · 供给产能/模型发布
Mythos is capable but the release has been withheld because they don't have the compute
tags: 好观点·好思考 · → daily
- <a id="atom-1c89810fdc16e303"></a>🟥 2026-04-28
X·@scaling01 · ECI score · 性能评估
Mythos should score 164 on ECI
tags: 好思考 · → daily
value: qty=164
- <a id="atom-8879093498e1423c"></a>🟥 2026-06-15
X·@scaling01 · AGI status · 技术路线
Mythos is not even close to AGI
tags: 好观点 · → daily
- <a id="atom-7df8609bcec2a137"></a>🟥 2026-06-09
X·@scaling01 · GPT-5.5 scaled up version could beat · 技术路线
Looks like a scaled up version of GPT-5.5 could beat Mythos 5
tags: 好观点 · → daily
📷 原图
- <a id="atom-c72bbe22f7d7cc2c"></a>🟥 2026-06-01
X·@zephyr_z9 · 训练数据量传闻估计 · 模型发布/训练数据
Rumors for mythos are around 150B-200B
tags: 好信源 · → daily
value: qty=150B-200B
- <a id="atom-ff4fcd3215bb3d74"></a>🟥 2026-06-01
X·@zephyr_z9 · 活跃参数范围 · 模型发布/技术路线
Active Param range around 200B
tags: 好信源 · → daily
value: qty=200B
- <a id="atom-f81238eed9bc69c9"></a>🟥 2026-06-01
X·@zephyr_z9 · 训练数据量 · 训练数据
Tokens around 200T
tags: 好信源 · → daily
value: qty=200T
- <a id="atom-c8bcaea2d936ff42"></a>🟥 2026-05-28
X·@scaling01 · forward looking gap on private benchmarks · 技术路线/模型发布
forward looking one is at least 12 months (at least for Mythos and GPT-5.5)
tags: 好观点 · → daily
value: qty=at least 12 months · direction=forward looking
📷 原图
- <a id="atom-ca2ad7def12453c5"></a>🟥 2026-05-26
X·@scaling01 · solved the unit distance problem · 数学问题
Mythos solving the unit distance problem is not surprising to me
tags: 好观点 · → daily
📷 原图
- <a id="atom-8589a6b1f0a8e476"></a>🟥 2026-05-26
X·@GaryMarcus · did not solve the unit distance problem · 数学问题
i don’t think mythos solved the problem
tags: 好观点 · → daily
- <a id="atom-76a8c7f908557057"></a>🟥 2026-05-26
X·@scaling01 · proof is right there · 数学问题
the proof is right there
tags: 好观点 · → daily
- <a id="atom-4c6d3612ea899dfb"></a>🟥 2026-05-22
X·@simobis23 · 存在性声明 · 模型发布
Mythos doesn't exist
tags: 好观点 · → daily
- <a id="atom-f0d4d876616357b5"></a>🟥 2026-05-22
X·@scaling01 · 存在性声明 · 模型发布
_yeah it's a hoax like the moon landing
dozens of companies are conspiring_
tags: 好观点 · → daily
- <a id="atom-93b2048a4351d3fc"></a>🟥 2026-05-07
X·@teortaxesTex · 模型能力相对于 gpt5.5 的表现 · 模型性能比较
I've talked to one access haver and his impression is that it's not noticeably stronger than 5.5
tags: 好信源 · → daily
- <a id="atom-12fb7c35ba86f3e9"></a>🟥 2026-07-01
X·@eli_lifland · 公共部署有助于社会反应 · 模型发布/监管政策
Public deployment helps societal response and isn't the most promising place to intervene
tags: 好思考 · → daily
📷 原图
- <a id="atom-96f07d817a5e0a1f"></a>🟥 2026-06-14
X·@verge · access by China · 地缘政治
China may have accessed Mythos
tags: 好思考 · → daily
- <a id="atom-ee8a686693383471"></a>🟥 2026-05-31
X·@gfodor · 有重大影响 · 技术路线/竞争格局
A lot riding on mythos
tags: 好思考 · → daily
🟦 新闻流 (squawk · 8)
展开新闻流 (FirstSquawk / financialjuice / wallstengine / DeItaone — 快讯, 非原创 take)
- <a id="atom-c761181227e3468e"></a>🟦 2026-06-21
X·@FirstSquawk · 突破NSA大部分机密网络的时间 · 安全事故
MYTHOS BREACHED NEARLY ALL CLASSIFIED SYSTEMS IN HOURS, NOT WEEKS
tags: 好信源 · → daily
value: date=2026-06-21 · qty=数小时内
- <a id="atom-a8c5f2c67fdf4cd0"></a>🟦 2026-06-16
X·@FirstSquawk · access status · 监管政策
Status of Mythos access remains as stated
tags: 好信源 · → daily
value: direction=unchanged
- <a id="atom-73faaa7e69cc2bc8"></a>🟦 2026-06-16
X·@FirstSquawk · operations verification · 监管政策
operations being verified
tags: 好信源 · → daily
value: direction=ongoing
- <a id="atom-420925c5f7bc1c36"></a>🟦 2026-06-02
X·@wallstengine ⭐ · vulnerabilities identified since launch · 产品发布/技术路线
Anthropic says Mythos has helped identify 10,000+ serious vulnerabilities since launch.
tags: 好数字·好信源 · → daily
value: qty=10,000+ · date=since launch
📷 原图
- <a id="atom-0eef8355f2c36261"></a>🟦 2026-05-11
X·@FirstSquawk [老化中] · AI 威胁评估 · 技术路线/监管政策
Japan’s Finance Minister Katayama noted concerns that advanced AI, including Mythos, could represent a major threat.
tags: 好信源 · → daily
value: direction=major threat
- <a id="atom-2d20bc0467074e72"></a>🟦 2026-04-30
X·@financialjuice · 被NSA用于识别网络漏洞 · 网络安全
NSA is using Mythos to identify cyber vulnerabilities.
tags: 好信源 · → daily
- <a id="atom-363b3457bfa6a3c2"></a>🟦 2026-04-30
X·@financialjuice · 白宫在考虑扩大工具接入权限 · 监管政策
NSA Uses Mythos to Hunt for Cyber Weaknesses as White House Weighs Broader Access
tags: 好信源 · → daily
- <a id="atom-e324b8256237cb14"></a>🟦 2026-04-30
X·@financialjuice · 不是末日设备 · 产品发布
White House Advisor Sacks: Mythos is not a doomsday device.
tags: 好观点 · → daily
⏱ 时间轴 (近 20)
- 🟥 2026-07-03 ·
narrative · potential positive outcomes of finding vulnerabilities in hu · → daily
- 🟥 2026-07-02 ·
narrative · comparison to Fable · → daily
- 🟥 2026-07-02 ·
narrative · is finding vulnerabilities in the human genome · → daily
- 🟥 2026-07-01 ·
narrative · 公共部署延迟 · → daily
- 🟥 2026-07-01 ·
narrative · 公共部署有助于社会反应 · → daily
- 🟦 2026-07-01 ·
fact · 被美国出口管制下线后恢复上线状态 · → daily
- 🟥 2026-06-28 ·
narrative · 被美政府用于战争 · → daily
- 🟦 2026-06-27 ·
fact · availability restricted to vetted US agencies and companies · → daily
- 🟥 2026-06-26 ·
narrative · 被禁的负面评价 · → daily
- 🟥 2026-06-26 ·
narrative · security performance relative to challengers · → daily
- 🟥 2026-06-26 ·
narrative · cost efficiency on specific benchmark · → daily
- 🟦 2026-06-26 ·
fact · 重新获批入境 · → daily
- 🟥 2026-06-25 ·
narrative · 被蒸馏与雇佣专家生成训练样本的差异 · → daily
- 🟥 2026-06-25 ·
narrative · AI licensing acceleration cause · → daily
- 🟥 2026-06-24 ·
narrative · 训练新模型优势 · → daily
- 🟥 2026-06-22 ·
position · multi-model routing performance does not match · → daily
- 🟥 2026-06-22 ·
forecast · 中国公司复现时间 · → daily
- 🟦 2026-06-21 ·
fact · 攻击速度 · → daily
- 🟦 2026-06-21 ·
fact · 突破NSA大部分机密网络的时间 · → daily
- 🟥 2026-06-21 ·
narrative · 可能发现运行在气隙系统上软件的漏洞 · → daily
← 实体目录 · 系统日志