Claude Opus 5.5（Anthropic） Claude Opus 5.5 (Anthropic)

来源Sources

为什么推荐Why we recommend it 同板 step-5-preview 的官方对照表写的是 Claude Opus 5（不是 5.5），互链只表示同板相关，不表示阶跃评过 Opus 5.5。社区绘画演示只在 HTML 库。整理质量 4 星（来源齐、边界清），不代表基准分。 On the vendor table Astra is ahead of Opus 5.5 on Terminal-Bench-Science 0.1 and AutomationBench, so the meta “leads” line is not treated as first on every row. The step-5-preview card’s vendor table names Claude Opus 5 (not 5.5); the cross-link is same-board only and does not mean StepFun evaluated Opus 5.5. The community painting demo stays in the HTML library.

要点（官方称 / 社区称）Highlights (vendor / community claims)

• 官方称：Claude 5.5 家族第一款；正文写多数工作达到 Fable 5.1 水平，典型负载比 Opus 5 少 40% 花费，输出比 Opus 5 快 30% 以上。同页又称 Sonnet 5.5 与 Haiku 5.5 将在随后数周推出。平台模型 id 写作 claude-opus-5-5，并称已在全平台可用，包括 Amazon Web Services、Google Cloud 与 Microsoft Azure。 Vendor claim: first model in the Claude 5.5 family; the page says it matches Fable 5.1 on most work, costs 40% less than Opus 5 on typical workloads, and generates output more than 30% faster than Opus 5. The same page says Sonnet 5.5 and Haiku 5.5 follow in the coming weeks. The platform model id is written claude-opus-5-5, and the page says it is available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure.
• 官方称（价目表，每百万 token）：cache reads $0.20（Opus 5 为 $0.50）、输入 $4（Opus 5 为 $5）、输出 $20（Opus 5 为 $25）、cache writes $5（Opus 5 为 $6.25）。页面另写输入/输出比 Opus 5 少 20%，cache reads 少 60%；fast mode（Claude Code 与 Claude Platform）最高 2.5 倍速度，输入 $8 / 输出 $40。本库未实测 API。 Vendor claim (price table, per 1M tokens): cache reads $0.20 (Opus 5 $0.50), input $4 (Opus 5 $5), output $20 (Opus 5 $25), cache writes $5 (Opus 5 $6.25). The page also says input/output are 20% less than Opus 5 and cache reads are 60% less; fast mode (Claude Code and the Claude Platform) is up to 2.5× speed at $8 input / $40 output. API not tested here.
• 官方称（对照表，列序 Opus 5.5 / Fable 5.1 / Opus 5 / GPT-6 Astra / GPT-5.6 Sol；页注除另有说明外 Opus 5.5 为 adaptive thinking、max effort；Terminal-Bench 4.0 的 Opus 5.5 为 xhigh、GPT-6 Astra 为 high，且为 OpenAI 所报）：Terminal-Bench 4.0 66.4% / 55.8% / 52.3% / 57.9% / 37.3%；FrontierCode v1.1 (Main) 54.4% / 50.3% / 48.0% / 53.3% / 47.5%。同表另有 Astra 更高的行：Terminal-Bench-Science 0.1 为 64.6% 对 Opus 5.5 的 58.7%；AutomationBench（页注由 Zapier 跑）为 41.4% 对 40.0%。页面自己也写：与 Fable 5.1 的实际差距比这些分数看起来更窄。本库未复跑。 Vendor claim (comparison table, columns Opus 5.5 / Fable 5.1 / Opus 5 / GPT-6 Astra / GPT-5.6 Sol; the page says Opus 5.5 results use adaptive thinking at max effort unless noted; Terminal-Bench 4.0 is xhigh for Opus 5.5 and high for GPT-6 Astra, the latter as reported by OpenAI): Terminal-Bench 4.0 66.4% / 55.8% / 52.3% / 57.9% / 37.3%; FrontierCode v1.1 (Main) 54.4% / 50.3% / 48.0% / 53.3% / 47.5%. The same table has rows where Astra is ahead: Terminal-Bench-Science 0.1 is 64.6% vs Opus 5.5 at 58.7%; AutomationBench (page note: run by Zapier) is 41.4% vs 40.0%. The page itself also says the real gap versus Fable 5.1 is narrower than these scores suggest. None re-run here.
• 第三方榜（Artificial Analysis，非官方、非本库）：页面标题为 “Claude Opus 5.5 (max with fallback)”，变体 “Adaptive Reasoning, Max Effort, Default Fallback”；Intelligence Index 58（#1 / 211），速度 Output tokens per second 为 N/A，价格输入 $4.00 / 输出 $20.00、Cache Discount 95%、Cost per Intelligence Index task $5.98，Verbosity 260M output tokens（#96 / 211）。技术规格写推理版、输入 text and image、输出 text、上下文 1M。发布月记为 September 2026。不把 AA 的 95% 缓存折扣与官方 $0.20 cache reads 合成新指标。 Third-party board (Artificial Analysis — neither vendor nor this library): page title “Claude Opus 5.5 (max with fallback)”, variant “Adaptive Reasoning, Max Effort, Default Fallback”; Intelligence Index 58 (#1 / 211); output tokens per second is N/A; pricing $4.00 input / $20.00 output, Cache Discount 95%, $5.98 cost per Intelligence Index task; verbosity 260M output tokens (#96 / 211). Specs list a reasoning version, text and image input, text output, and a 1M context. Release month is written September 2026. The AA 95% cache discount is not merged here with the vendor’s $0.20 cache-read price into a new figure.
• 社区演示（echohive 页面自述，非本库复跑，也不另开 Models demo 卡）：https://www.echohive.ai/experiments/dinosaurs 《The Last Day》自称 “A 2.5D PAINTING · MADE BY OPUS 5.5”。条目在 HTML 库 last-day-dinosaurs，不进本板。 Community demo (echohive page self-description, not re-run here, and not a separate Models demo card): https://www.echohive.ai/experiments/dinosaurs “The Last Day” calls itself “A 2.5D PAINTING · MADE BY OPUS 5.5”. The entry lives in the HTML library as last-day-dinosaurs, not on this board.