🕷️
SetupAI
💰較早收集於 9h

Anthropic 在企業客戶採用率上超越 OpenAI

Anthropic 在企業客戶採用率上超越 OpenAI
PostLinkedIn
💰閱讀原文： TechCrunch AI
#market-share#enterprise-ai#llm-adoptionanthropicanthropicopenairamp

💡Anthropic 的企業採用率正式超越 OpenAI，是時候重新評估您的 LLM 供應商策略了。

⚡ 30-Second TL;DR

有什麼變化

根據 Ramp 數據，Anthropic 在企業客戶市佔率上已領先 OpenAI。

為什麼重要

此轉變顯示企業 AI 策略正趨向多模型化，降低對單一供應商的依賴。這也標誌著市場競爭加劇，Anthropic 對安全性和特定模型能力的專注正獲得企業買家的青睞。

下一步行動

評估您目前的 LLM 技術堆疊，並對 Claude 3.5 Sonnet 與 GPT-4o 進行基準測試，確認 Anthropic 的產品是否更適合您的企業應用場景。

誰應關注：Enterprise & Security Teams

關鍵要點

  • •根據 Ramp 數據，Anthropic 在企業客戶市佔率上已領先 OpenAI。
  • •34.4% 的受訪企業付費使用 Anthropic，而 OpenAI 為 32.3%。
  • •此趨勢顯示企業在選擇 LLM 供應商時，正逐漸轉向多元化，不再僅依賴市場龍頭。

🧠 深度解析

Web-grounded analysis with 29 cited sources.

🔑 增強重點摘要

  • •Anthropic's business customer adoption quadrupled over the past year, while OpenAI's grew by only 0.3% in the same period, marking a significant shift from April 2025 when OpenAI commanded roughly 32% of business AI adoption compared to Anthropic's under 8%.
  • •Anthropic is currently winning approximately 70% of head-to-head matchups against OpenAI among businesses making their initial AI service purchases, indicating a reversal of 2025 trends where OpenAI adoption accelerated faster.
  • •The surge in Anthropic's adoption is largely attributed to its enterprise-focused tools like Claude Code and Cowork, which cater to developer workflows and have gained substantial traction in corporate procurement.
  • •The Ramp AI Index, which reported this shift, is based on corporate card and bill-pay data from over 50,000 U.S. businesses, counting paid transactions, though it may underestimate usage of free AI tools or personal accounts.
  • •Despite Anthropic's growth, a significant overlap exists, with 79% of companies paying for Anthropic also subscribing to OpenAI, suggesting a multi-vendor strategy rather than exclusive adoption.
📊 競品分析▸ Show

AI Enterprise Platform Comparison: Anthropic Claude vs. OpenAI ChatGPT (as of May 2026)

Feature / ProviderAnthropic Claude EnterpriseOpenAI ChatGPT Enterprise
Primary ModelsClaude Opus 4.6/4.7, Sonnet 4.6, Haiku 4.5, Claude Code, Claude CoworkGPT-5.5, GPT-4o, ChatGPT Team, ChatGPT Enterprise, Codex
Enterprise PricingCustom pricing; Self-serve Enterprise starts at $20/seat/month (billed annually, minimum 20 seats). Team plan: $25/user/month (billed annually, minimum 5 users).Custom pricing; Suggested $60-$100+/seat/month (depending on volume/features, annual commitment). ChatGPT Team: $20/user/month (billed annually, 2+ users).
Context WindowUp to 1M tokens (Claude Opus 4.6/4.7, Mythos Preview); Claude Team: 200K tokens; Claude Enterprise (Sonnet 4.6 chat): 500K tokens.GPT-4o/ChatGPT Enterprise: 128K tokens; GPT-5.5 (reasoning): 196K tokens; GPT-5.5 (non-reasoning): 32K tokens.
Coding CapabilitiesClaude Code (CLI-native autonomous coding agent); Opus 4.6 achieved 80.8% on SWE-Bench Verified.Codex (developer coding tool); ChatGPT Team includes Advanced Data Analysis.
Multimodal SupportClaude Opus 4.7 can understand text and image inputs, output text, diagrams, and audio via text-to-speech.Broad multi-modal capabilities including text, code, vision, and voice; DALL-E image generation.
Safety & AlignmentConstitutional AI (training with natural language principles for helpful, honest, harmless behavior).Focus on enterprise-grade security and privacy; data not used for training.
Enterprise FeaturesCentralized administration, collaboration features (Projects, Cowork), SSO, audit logging, custom data retention, compliance APIs, SCIM.Unlimited high-speed access, advanced data analysis, custom GPTs, plugins, dedicated admin console, SSO, domain verification, analytics, API credits, Microsoft 365 integration.
Data PrivacyDoes not train models on customer conversations and content.Does not train on business data.
Market FocusPrimarily enterprise-focused, leveraging early adopter base to go mainstream.Strong consumer default, now expanding aggressively into enterprise.

🛠️ 技術深入

  • Constitutional AI (CAI): Anthropic's core approach to AI safety, training models to be helpful, honest, and harmless without extensive human feedback for harmlessness. It involves two stages: a supervised learning stage where the AI critiques and revises its own responses based on a set of natural language principles (a 'constitution'), and a reinforcement learning stage where the AI learns from its self-generated feedback.
  • Model Architecture: Claude models are built upon a decoder-only Transformer architecture, a neural network design that uses self-attention mechanisms to weigh relationships between tokens across the entire input sequence. This foundation is augmented with training pipelines, alignment layers, long-context engineering, and safety guardrails.
  • Safety Layers: Claude's safety architecture is layered, incorporating input filtering to identify harmful content or prompt injections, output moderation to check for policy violations, and policy models with refusal heuristics to reduce risks across the request lifecycle.
  • Context Window: Claude 3 models can process up to 200,000 tokens in a single request. More advanced models like Claude Opus 4.6, Opus 4.7, and Mythos Preview offer an extended context window of up to 1 million tokens, crucial for analyzing lengthy documents and complex codebases.
  • Claude Code Architecture: This autonomous coding agent employs a single-threaded master loop (codenamed 'nO') designed for debuggability, transparency, and reliability. It utilizes a persistent sandboxed REPL (Read-Eval-Print Loop) and filesystem to ensure verifiable code execution, where the model decides what to compute, and the grounding layer executes it.
  • Training Infrastructure: Anthropic leverages cloud computing resources from Amazon Web Services (AWS) and Google Cloud Platform (GCP) for training its AI systems, utilizing development frameworks such as PyTorch, JAX, and Triton.

🔮 前景展望AI analysis grounded in cited sources

Anthropic's market lead in business adoption may be precarious.
The same Ramp report that highlights Anthropic's lead also warns that its position is threatened by escalating costs, compute constraints, and its token-based pricing model, which has already led to budget overruns for some large enterprises.
Enterprises will increasingly adopt a multi-model AI strategy.
The high overlap of businesses using both Anthropic and OpenAI (79% of Anthropic users also pay for OpenAI) suggests that companies are selecting specific AI tools based on use case rather than standardizing on a single provider.
The focus on specialized, developer-centric AI tools will intensify competition.
Anthropic's success is significantly driven by its enterprise tools like Claude Code and Cowork, indicating that specialized AI solutions tailored for specific workflows, particularly in development, are key drivers for enterprise adoption.

⏳ 時間線

2021-01
Anthropic founded by former OpenAI executives Dario and Daniela Amodei as a Public Benefit Corporation.
2022-12
Anthropic publishes its foundational paper on 'Constitutional AI: Harmlessness from AI Feedback'.
2023-03
Anthropic publicly launches its AI assistant, Claude.
2023-09
Amazon invests $4 billion in Anthropic, forming a strategic partnership.
2024-03
Anthropic launches the Claude 3 family of models (Haiku, Sonnet, Opus).
2026-04
Anthropic surpasses OpenAI in business customer adoption, according to the Ramp AI Index.
💰閱讀原文： TechCrunch AI
📰

AI 週報

閱讀本週精選 AI 大事摘要 →

👉相關動態

同主題

探索 #market-share

同產品

更多關於 anthropic

同來源

最新來自 TechCrunch AI

Sam Altman 與 AI 發展速度的「減速」辯論

Sam Altman 與 AI 發展速度的「減速」辯論

TechCrunch AI•Aug 2

AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： TechCrunch AI ↗