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
Anthropic Opus 4.8
19 atoms · 跨 11 天 · 首见 2026-05-28 · 最近 2026-07-01
三色: 🟦 fact 9 · 🟥 take 10 · stance ▲4/▼9/◆3
来源: X 19
时态: fresh:17 · stale:2
标签: 好观点:10 · 好数字:6 · 好信源:4 · 好思考:1
叙事 (narrative) (8)
- 2026-06-17 · security status conclusion
the reasonable conclusion is not that frontier models are safe, but that even the best, most-tested frontier models remain reliably breakable under sustained automated pressure
tags: 好观点·好思考 · → daily
value: qty= · date=
- 2026-05-28 · AAI 排名
Opus 4.8 takes the top spot on AAI
tags: 好信源 · → daily
value: qty=top · date=2026-05-28
- 2026-07-01 · tends to over-simplify complex theories
I have to use various forms of mental jiu-jitsu to get Op. 4.8 to stop quenching the theory to some oversimplification.
tags: 好观点 · → daily
- 2026-06-28 · 推理能力显著下降
First time I've seen this happen (and I use it daily): it almost totally stopped thinking, even at highest reasoning setting. And answered terribly.
tags: 好观点 · → daily
- 2026-05-31 · 模型质量可能被低估
It is possible that Opus 4.8 is a much better model than I give it credit for, and it is held back terribly by Claude Code.
tags: 好观点 · → daily
- 2026-05-28 · 模型进步幅度
the gap between current models and the next one (4.7 -> 4.8) won’t show a dramatic jump
tags: 好观点 · → daily
- 2026-05-28 · 模型发布性质
Mostly hype. Just PR and who has the biggest.
tags: 好观点 · → daily
- 2026-05-28 · 性能提升
Opus 4.8 is smarter/faster and has real Excel taste
tags: 好观点 · → daily
value: direction=优于 4.7
立场 (position) (3)
- 2026-06-30 · 性能评价
After using Fable 5, Opus 4.8 seems not strong for me
tags: 好观点 · → daily
value: direction=弱于
展开 2 条老旧 / 已过期
- 2026-06-03
[老旧] · 产品质量评价
Opus 4.8太垃圾了，而且废话还多，一股子ChatGPT味
tags: 好观点 · → daily
- 2026-06-02
[老旧] · 使用体验评价
Gave 4.8 another try, regardless of how you tweak it, it’s trash
tags: 好观点 · → daily
事实 (fact) (8)
- 2026-06-17 · worst-case attack success rate
The strongest adaptive search (tree-of-attacks) breaks Opus 4.8 on 11.5% of intents overall
tags: 好数字·好信源 · → daily
value: qty=11.5% · date=
- 2026-06-19 · launch date
Opus 4.8 was launched on May 28, 2026
tags: 好数字 · → daily
value: date=2026-05-28
- 2026-06-17 · 최강 공격(TAP) 성공률
Opus 4.8: 최강 공격(TAP)에서 11.5% 성공
tags: 好数字 · → daily
value: qty=11.5% · date=2026-06-17
- 2026-06-17 · adversarial robustness evaluation result
researchers evaluated the adversarial robustness of two frontier LLMs developed by Anthropic, Fable 5 and Opus 4.8
tags: 好信源 · → daily
value: qty= · date=
- 2026-06-17 · jailbreak resistance result
both models resist the majority of attacks, but the residual surface is larger than aggregate framing suggests: it is dominated by adaptive iterative attacks, while static obfuscation is near-fully neutralized
tags: 好信源 · → daily
value: qty= · date=
- 2026-06-15 · FACTS-Parametric评测得分
Opus 4.8 25.1%.
tags: 好数字 · → daily
value: qty=25.1%
- 2026-06-01 · 状态
Anthropic Opus 4.8 is new SOTA on ARC-AGI-3 Score: 1.5%
tags: 好数字 · → daily
value: qty=1.5% · score=SOTA on ARC-AGI-3
- 2026-05-28 · cost reduction
3x cheaper
tags: 好数字 · → daily
value: qty=3x cheaper
时间轴 (近 20)
- 2026-07-01 ·
narrative · tends to over-simplify complex theories · → daily
- 2026-06-30 ·
position · 性能评价 · → daily
- 2026-06-28 ·
narrative · 推理能力显著下降 · → daily
- 2026-06-19 ·
fact · launch date · → daily
- 2026-06-17 ·
fact · 최강 공격(TAP) 성공률 · → daily
- 2026-06-17 ·
fact · adversarial robustness evaluation result · → daily
- 2026-06-17 ·
fact · jailbreak resistance result · → daily
- 2026-06-17 ·
fact · worst-case attack success rate · → daily
- 2026-06-17 ·
narrative · security status conclusion · → daily
- 2026-06-15 ·
fact · FACTS-Parametric评测得分 · → daily
- 2026-06-03 ·
position · 产品质量评价 · → daily
- 2026-06-02 ·
position · 使用体验评价 · → daily
- 2026-06-01 ·
fact · 状态 · → daily
- 2026-05-31 ·
narrative · 模型质量可能被低估 · → daily
- 2026-05-28 ·
narrative · 模型进步幅度 · → daily
- 2026-05-28 ·
narrative · 模型发布性质 · → daily
- 2026-05-28 ·
narrative · 性能提升 · → daily
- 2026-05-28 ·
narrative · AAI 排名 · → daily
- 2026-05-28 ·
fact · cost reduction · → daily
← 实体目录 · 系统日志