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
Harvey
37 atoms · 跨 14 天 · 首见 2026-05-02 · 最近 2026-07-01
三色: 🟦 fact 20 · 🟥 take 17 · stance ▲11/▼7/◆7
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: X 37
时态: fresh:34 · stale:2 · aging:1
标签: 好观点:11 · 好信源:10 · 好思考:9 · 好数字:8 · 好问题:1
🟨 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
Harvey 在 Q2 2026 实现了 +$100M NNARR 和 53% DAU/MAU，每周处理 54TB 文档
¹¹¹。产品层面，团队统一了所有工具入口，用户通过 agent 学习发现更多功能
¹¹。多智能体研究证明其方案能以更低成本运行
¹，同时 Harvey 正在挑战 Thomson Reuters 等传统法律科技巨头
¹。但质疑者认为 Harvey 的数据优势不构成护城河，甚至不如 Kirkland & Ellis 等 BigLaw 律所
¹¹。更悲观的观点指出，若 OpenAI 和 Anthropic 推出通用模型，Harvey 可能被挤压至出售或消亡
¹。另有猜测 Harvey 实际可能已混用 DeepSeek 等廉价模型来降低成本
¹。综合来看，Harvey 在产品执行力与增长指标上强劲，但数据壁垒和底层模型依赖仍是结构性分歧焦点
¹¹。
🧭 拥挤度 (一人一票): ▲ 4 位作者 (KOL4) vs ▼ 3 位作者 (KOL3)
📄 广泛报道的事实 · 1 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)
- 与Applied Compute合作训练法律AI助手 — 2 源报道 ·
X
🟦 客观事实 (facts) (20)
- <a id="atom-b6defaf06985a0ad"></a>🟦 2026-05-27
X·@winstonweinberg · 闭源前沿模型完成每个 LAB 任务平均耗时 22 分钟 · 成本
The best-performing closed-source frontier models took an average of 22 minutes to complete each LAB task
tags: 好数字·好信源 · → daily
value: qty=22 minutes · date=na
📷 原图
- <a id="atom-a85dee280f674491"></a>🟦 2026-05-27
X·@winstonweinberg · 闭源前沿模型每个 LAB 任务平均推理成本 50 美元 · 成本
with $50 per-task average inference cost.
tags: 好数字·好信源 · → daily
value: qty=$50 · date=na
📷 原图
- <a id="atom-4919c4eb858dbd31"></a>🟦 2026-07-01
X·@gabepereyra · Q2 2026 NNARR · 营收
+$100M NNARR
tags: 好数字 · → daily
value: qty=+$100M · date=Q2 2026
📷 原图
- <a id="atom-af8c4bb9aabf1a6a"></a>🟦 2026-07-01
X·@gabepereyra · Q2 2026 DAU/MAU · 出货量
53% DAU/MAU
tags: 好数字 · → daily
value: qty=53% · date=Q2 2026
📷 原图
- <a id="atom-58d2ccf192f4af3e"></a>🟦 2026-07-01
X·@gabepereyra · 文档处理规模 · 供给产能
Scaling document processing (54TB / week)
tags: 好数字 · → daily
value: qty=54TB / week
📷 原图
- <a id="atom-ddbfb9540bb79692"></a>🟦 2026-07-01
X·@gabepereyra · 云代理基础设施投资 · 资本开支/技术路线
We invested heavily in cloud agent infrastructure at the end of last year and in Q1
tags: 好思考 · → daily
value: date=end of last year and Q1
📷 原图
- <a id="atom-f4be78d57ccd4cec"></a>🟦 2026-07-01
X·@gabepereyra · 产品表面统一化 · 产品发布
In Q2 we also unified many of our product surfaces (collapsed as @winstonweinberg says) by making them all tools accessible by our cloud agents
tags: 好思考 · → daily
value: date=Q2
📷 原图
- <a id="atom-2f2c9d0e94c0859f"></a>🟦 2026-07-01
X·@gabepereyra · 用户发现产品功能趋势 · 用户渗透
we are seeing users discover more of the product by learning from their agents (see plot of product surfaces per user)
tags: 好思考 · → daily
value: date=Q2 2026
📷 原图
- <a id="atom-3233cd7a4419ecc0"></a>🟦 2026-06-22
X·@gabepereyra · 与Applied Compute合作训练法律AI助手 · 合作客户
Harvey partnered with @appliedcompute to train a legal agent.
tags: 好信源 · → daily
- <a id="atom-25764a9ecb308689"></a>🟦 2026-06-05
X·@deedydas · 估值 · 估值
Harvey - $11B
tags: 好数字 · → daily
value: qty=$11B
- <a id="atom-6192f28afa089abd"></a>🟦 2026-05-29
X·@theinformation · growing fast · 合作客户
Harvey and Legora are still growing fast despite fears that frontier AI models would undercut legal AI apps.
tags: 好信源 · → daily
- <a id="atom-6dbdfb0946367350"></a>🟦 2026-05-27
X·@winstonweinberg · 发布与 Baseten 的开源权重法律智能体研究合作 · 产品发布/合作客户
Today we're sharing our first research collaboration with @baseten on open-weight legal agents.
tags: 好信源 · → daily
📷 原图
- <a id="atom-371e6ac205d54fa8"></a>🟦 2026-05-27
X·@winstonweinberg · 使用 LAB 基准后训练开源权重模型以匹配闭源前沿模型性能 · 产品发布/技术路线
Using signal from LAB (our Legal Agent Benchmark of 1,200+ tasks across 24 practice areas), we post-trained an open-weight model to match closed-source frontier performance.
tags: 好信源 · → daily
📷 原图
- <a id="atom-091a4c5354921bd4"></a>🟦 2026-05-27
X·@winstonweinberg · 闭源基础模型提供者避免通过 API 暴露原始推理 token 以防止模型蒸馏 · 技术路线
Closed-source foundation model providers avoid exposing raw reasoning tokens via API to prevent model distillation.
tags: 好信源 · → daily
📷 原图
- <a id="atom-775f07cce16bd192"></a>🟦 2026-05-19
X·@nayakkayak · 员工人数 · 人事变动
we were just 20 people
tags: 好数字 · → daily
value: qty=20
📷 原图
- <a id="atom-d71bbf9f514be32c"></a>🟦 2026-05-19
X·@nayakkayak · 客户数 · 合作客户
5 customers
tags: 好数字 · → daily
value: qty=5
📷 原图
- <a id="atom-e8ba75a7515d48cd"></a>🟦 2026-06-17
X·@gabepereyra · 模型策略 · 技术路线/产品发布
_Model strategy for @harvey:
We are working on the first model in our legal foundation model series, inspired by @cursor_ai's Composer._
tags: 好信源 · → daily
- <a id="atom-68fb913530a631b5"></a>🟦 2026-05-05
X·@harvey · 发布500+法律AI代理及Agent Builder · 产品发布
Introducing 500+ legal agents and a new Agent Builder in Harvey.
tags: 好信源 · → daily
value: date=2026-05-05
📷 原图
- <a id="atom-e29b3e94a56c787b"></a>🟦 2026-06-22
X·@winstonweinberg · 与Applied Compute合作训练法律代理 · 合作客户
Harvey partnered with @appliedcompute to train a legal agent.
tags: 好思考 · → daily
- <a id="atom-59a7fc72f6989c43"></a>🟦 2026-06-02
X·@reissbaker · cannot train on client data due to attorney/client privilege restrictions · 监管政策/数据壁垒
harvey is not, so it can't actually train on anything
tags: 好思考 · → daily
🟥 多头 takes (bullish) (5)
- <a id="atom-e08d91bafc3636c1"></a>🟥 2026-06-13
X·@ethankongee · 多智能体研究结果 · 技术路线/成本效率
Harvey has done a research on this with GLM 5.1 advised by Opus 4.7 and proved that it works at much lower cost.
tags: 好信源 · → daily
value: direction=works at much lower cost
- <a id="atom-6c81f4b60a58bf98"></a>🟥 2026-06-04
X·@Techmeme · 挑战法律科技行业现有企业 · 竞争格局
How AI tools from Harvey, Legora, and Anthropic are challenging legal tech incumbents like Thomson Reuters and LexisNexis, which are upgrading their products
tags: 好观点 · → daily
- <a id="atom-42e2c64497676e12"></a>🟥 2026-06-05
X·@gabepereyra · 产品策略：构建协作平台，让律所和客户安全共享数据并训练定制模型 · 产品发布/合作客户
We’re building a collaborative platform (shared spaces) that allows law firms and clients to securely share data on client matters and build custom agents for their clients in a secure way.
tags: 好观点 · → daily
- <a id="atom-fc2190a9efdf9cb6"></a>🟥 2026-05-19
X·@nayakkayak [老旧] · 经营状态 · 管理层表态
I remain incredibly bullish in @winstonweinberg, @gabepereyra, @aniqued, & @sgurumur's leadership to take the company to the next level
tags: 好观点 · → daily
📷 原图
- <a id="atom-ebf36265413d24d7"></a>🟥 2026-05-19
X·@nayakkayak · 估值标签 · 技术路线
three AI decacorns at three different layers of the stack: Harvey (apps)
tags: 好观点 · → daily
📷 原图
🟥 空头 takes (bearish) (7)
- <a id="atom-80a34a2392edaec1"></a>🟥 2026-06-02
X·@reissbaker · data unlikely to be a moat · 数据壁垒
harvey's data, to the extent they have any, is unlikely to be a moat
tags: 好思考 · → daily
- <a id="atom-6827e1cdc85778e6"></a>🟥 2026-06-02
X·@reissbaker · does not have data advantage over BigLaw firms · 数据壁垒/竞争格局
worse for harvey is that they most likely do not even have a data advantage over BigLaw firms at all
tags: 好思考 · → daily
- <a id="atom-3875adeda5a9b35d"></a>🟥 2026-06-02
X·@reissbaker · positioned worse on data advantage than Kirkland & Ellis · 数据壁垒/竞争格局
harvey has no data moat over kirkland & ellis, and in fact is worse-positioned from a data perspective than the firm is
tags: 好思考 · → daily
- <a id="atom-ece391df949ea27b"></a>🟥 2026-06-02
X·@reissbaker · squeezed by OpenAI and Anthropic if generalized models sweep · 竞争格局/产品命运
openai and anthropic will squeeze harvey like a lime on light mexican beer, and harvey will either sell or die
tags: 好观点 · → daily
- <a id="atom-f5ea563216d36c75"></a>🟥 2026-05-06
X·@NadimHossain [老旧] · 业务存续状态 · 业务模式
I'm bearish on Legora/Harvey precisely because they exist only in the temporary state
tags: 好观点 · → daily
- <a id="atom-ee1addf358d81e47"></a>🟥 2026-05-06
X·@NadimHossain · 未来归零风险 · 业务模式
They will probably go to zero.
tags: 好观点 · → daily
- <a id="atom-c14bcf533f5e4afa"></a>🟥 2026-05-02
X·@0xdoug [老化中] · 可能实际使用更便宜模型 · 成本控制
Is Harvey really pumping every single token through Opus? Or are they carefully measuring what can run in DeepSeek Flash at 95% lower cost?
tags: 好问题 · → daily
value: direction=负面
🟥 中性 takes (neutral) (2)
- <a id="atom-339b424d5b305f3e"></a>🟥 2026-06-03
X·@reissbaker · would be crazy to violate contracts with 100% lawyer and law firm customers · 监管政策/数据壁垒
given their customer base (100% lawyers and law firms) imo they'd be crazy to violate their contracts
tags: 好观点 · → daily
- <a id="atom-377f65e2c195c943"></a>🟥 2026-05-06
X·@NadimHossain · 被AI实验室收购可能性 · 并购
Unless an AI lab buys them for training data.
tags: 好观点 · → daily
⏱ 时间轴 (近 20)
- 🟦 2026-07-01 ·
fact · Q2 2026 NNARR · → daily
- 🟦 2026-07-01 ·
fact · Q2 2026 DAU/MAU · → daily
- 🟦 2026-07-01 ·
fact · 文档处理规模 · → daily
- 🟦 2026-07-01 ·
fact · 云代理基础设施投资 · → daily
- 🟦 2026-07-01 ·
fact · 产品表面统一化 · → daily
- 🟦 2026-07-01 ·
fact · 用户发现产品功能趋势 · → daily
- 🟦 2026-06-22 ·
fact · 与Applied Compute合作训练法律AI助手 · → daily
- 🟦 2026-06-22 ·
fact · 与Applied Compute合作训练法律代理 · → daily
- 🟦 2026-06-17 ·
fact · 模型策略 · → daily
- 🟥 2026-06-13 ·
narrative · 多智能体研究结果 · → daily
- 🟥 2026-06-05 ·
narrative · 产品策略：构建协作平台，让律所和客户安全共享数据并训练定制模型 · → daily
- 🟦 2026-06-05 ·
fact · 估值 · → daily
- 🟥 2026-06-04 ·
narrative · 挑战法律科技行业现有企业 · → daily
- 🟥 2026-06-03 ·
narrative · would be crazy to violate contracts with 100% lawyer and law · → daily
- 🟥 2026-06-02 ·
narrative · data unlikely to be a moat · → daily
- 🟥 2026-06-02 ·
narrative · does not have data advantage over BigLaw firms · → daily
- 🟥 2026-06-02 ·
forecast · squeezed by OpenAI and Anthropic if generalized models sweep · → daily
- 🟥 2026-06-02 ·
narrative · positioned worse on data advantage than Kirkland & Ellis · → daily
- 🟦 2026-06-02 ·
fact · cannot train on client data due to attorney/client privilege · → daily
- 🟦 2026-05-29 ·
fact · growing fast · → daily
← 实体目录 · 系统日志
