Claude Opus 5.5 Tops AA Intelligence Index, Faster and Cheaper in Early Tests
Anthropic 于 9 月 23 日发布 Claude Opus 5.5，新模型在 Artificial Analysis 智能指数（AAII）上登顶并以较大优势领先前 SoTA，官方同步降价约 20% 并加大缓存命中折扣。多位独立开发者与研究者的第一手实测普遍积极，性价比成为核心卖点，但也有人提醒 Opus 5 当年「跑分强、体验翻车」的前车之鉴，实际口碑仍需时间检验。
已确认
- Opus 5.5 在 Artificial Analysis 智能指数登顶。据 AAII 榜单，其综合智能分大幅领先；爆料称 max with fallback 版本拿到 58 分，比此前纪录（Fable 5.1 与 GPT-6 Astra）高约 5 分；在 xhigh 推理档位下得分超过 Fable 5.1 Max 且 token 用量更少（m2/m3/m7/m10/m11）
- Anthropic 将价格下调约 20%，并加大缓存命中折扣（m3）；发布消息称比 Opus 5 便宜最高 40%（m20）
- 多位独立实测者给出一致体验：开发者 Lydia Hallie 与 AI 研究者 Yuchen Jin 均称每任务比 Opus 5 快约 30%、便宜约 40%；Lydia Hallie 表示整体体验更像 Fable，Yuchen Jin 称要重新捡起搁置一个月的 Claude Code（m4/m13）
- Bindu Reddy 实测称模型「一发命中」完成几乎所有任务，是最快的一次通过方式，价格低于上一代 Opus（m5/m9）
- 编程实测亮点：审计并修复 20 万行代码库用时不到 3 小时，Opus 5 需 20+ 小时，token 消耗减少 2.5 倍；有早期测试者交付 68 万行代码迁移；中文博主 vista8 汇总成本约为 GPT-6 Astra 的五分之一，基准成绩 Terminal-Bench 4.0 66.4%、FrontierCode v1.1 54.4%、CursorBench 4.0 57.8%（m16/m18）
- 第三方评测称其在 agentic coding、知识工作、推理、computer use、视觉识别等测试项上全面超越 Fable 5.1 与 GPT-6 Astra，覆盖发布仅三周的竞品；Nous Research 已将其接入 Nous Portal 的 Hermes Agent（m8/m14/m17）
- 用户体验方面：有用户称价格约为前代 Fable 的 1/3，在 Max 订阅下额度大幅提升；抢先体验者称交互手感延续 Opus 4.6 的顺畅风格；Every 团队实测后观察到部分 Codex 用户回流（m15/m19/m20）
尚未确认
- 降价幅度存在多种口径：官方降价约 20%（m3）、比 Opus 5 便宜 25%（bindureddy，m5）与最高 40%（官方发布消息，m20）并存，「40%」多为每任务实际成本下降的实测结论
- m8、m11、m14、m15 均为转发或爆料性质，基准细节尚待官方验证
为什么重要
- RL 研究者 Nat Lambert 指出，Opus 5.5 是所有被测模型中每任务消耗 token 最多的「token 吞噬怪兽」，这或许正是 Anthropic 降价的原因——榜单登顶反而可能是红旗（m12）
- Angaisb 提醒，Opus 5 当年跑分也曾高于 Fable 5，但用户实际体验不买账；Opus 5 发布约 48 小时后即暴露不少毛病的前例让部分用户持观望态度（m6/m19）
- 若实测口碑持续，Opus 5.5 可能改变 agentic coding 与知识工作工具的竞争格局，甚至影响用户对 GPT-6 系列后续模型的预期（m6/m7/m20）
2026-09-23 ~ 2026-09-23 · 22 related posts
- Episode 1: Claude Opus 5.5 Spotted in Claude Code Ahead of Launch(2026-09-23, 3 posts)
- Episode 2: Anthropic Launches Claude Opus 5.5: Smarter, Faster and 40% Cheaper(2026-09-23, 56 posts)
- Episode 3: Claude Opus 5.5 Tops AA Intelligence Index, Faster and Cheaper in Early Tests(2026-09-23, 22 posts)
- Episode 4: Anthropic Launches Opus 5.5, Leading in Agentic Coding and Computer Use(2026-09-23, 5 posts)
- Episode 5: Anthropic Teases Sonnet 5.5 and Haiku 5.5 Within Weeks(2026-09-23, 2 posts)
Primary sources
- Claude Opus 5.5 tops Artificial Analysis Intelligence Index with 20% price cut — minxio_ ·
- Hands-on with Opus 5.5: feels like Fable, ~30% faster and ~40% cheaper per task than Opus 5 — lydiahallie ·
- Opus 5.5 audits a 200k-line codebase in under 3 hours using 2.5x fewer tokens than Opus 5 — minchoi ·
- Opus 5.5 benchmarks: beats Fable 5.1 and GPT 6 Astra at half the price — Scobleizer · 2026-09-23
- Opus 5.5 posts strong benchmarks, but the Opus 5 precedent looms over GPT-6 Sol fears — Angaisb_ · 2026-09-23
- Hands-on With Opus 5.5: Same Feel as 4.6, More Power, ~40% Cheaper Than Opus 5 — EricBuess · 2026-09-23
- [source] Hands-on with Opus 5.5: feels like Fable, ~30% faster and ~40% cheaper per task than Opus 5 — lydiahallie · 2026-09-23
- [source] Claude Opus 5.5 tops Artificial Analysis Intelligence Index with 20% price cut — minxio_ · 2026-09-23
- Dev Praises Claude Opus 5.5: One-Shot Success, Faster and Cheaper Than Opus 5 — bindureddy · 2026-09-23
- Claude Opus 5.5 Tops AI Index at 58, Cuts Price 20% to Match GPT-6 Astra — brandon_galang · 2026-09-23
- Opus 5.5 is ~30% faster and ~40% cheaper per task than Opus 5, researcher says — Yuchenj_UW · 2026-09-23
- Opus 5.5 Leads Artificial Analysis Intelligence Index by a Wide Margin — Angaisb_ · 2026-09-23
- Opus 5.5 Jumps to 1st Place on the AA Intelligence Index — UnknownEssence · 2026-09-23
- Opus 5.5 Lauded as One-Shot Machine, 25% Cheaper Than Opus 5 — bindureddy · 2026-09-23
- Claude Opus 5.5 lands in Hermes Agent, claims sweep of Fable 5.1 and GPT 6 Astra at half the price — NousResearch · 2026-09-23
- Opus 5.5 is 40% cheaper and 30% faster than Opus 5, says Reddit post — Seeker_Of_Knowledge2 · 2026-09-23
- Opus 5.5 ships ~30% faster and ~40% cheaper per task than Opus 5 — haider1 · 2026-09-23
- AAII ranking: Opus 5.5 tops overall intelligence, beats Fable 5.1 Max with fewer tokens at xhigh — Angaisb_ · 2026-09-23
- Leak: Claude Opus 5.5 reportedly scores 58 on AA Intelligence Index, 5 points over SoTA — airesearch12 · 2026-09-23
- Nat Lambert flags Opus as top token-gobbler: leading the Intelligence Index is a red flag — natolambert · 2026-09-23
- Anthropic ships Opus 5.5: up to 40% cheaper than Opus 5, pulling Codex users back to Claude — danshipper · 2026-09-23
- Claude Opus 5.5 dominates benchmarks of rivals released three weeks ago — daniel_mac8 · 2026-09-23
- Opus 5.5 field results: 200k-line codebase fixed in under 3 hours at 1/5 the cost of GPT-6 Astra — vista8 · 2026-09-23
- [source] Opus 5.5 audits a 200k-line codebase in under 3 hours using 2.5x fewer tokens than Opus 5 — minchoi · 2026-09-23
- Early users: Opus 5.5 is ~3x cheaper than Fable with much higher Max usage — doodlestein · 2026-09-23