Leaked OpenAI Memo: Project Spud Competes with Mythos in Direct Challenge to Anthropic

An internal memo from OpenAI, reportedly leaked, reveals a strategic plan for the second quarter of 2026 drafted by Chief Revenue Officer Denis D'Almeida. The document indicates that OpenAI is preparing to compete with Anthropic through technological upgrades and ecosystem partnerships.

Project Spud: The Next-Generation Reasoning Model

The memo introduces a new AI model under the codename Spud (likely GPT-5o or GPT-5.5), explicitly positioned as the main tool to rival Anthropic's Claude Mythos.

Blackwell Architecture: The model was trained using NVIDIA's Blackwell chips.

Performance Breakthrough: Early feedback from customers shows that Spud performs better in complex reasoning, intent recognition, and reliability in production settings.

SuperApp Vision: OpenAI intends to integrate this model into its core products, moving toward a "SuperApp" format.

Frontier Platform: Setting the Standard for Enterprise Agents

OpenAI is building a platform called Frontier, designed to become the default foundation for enterprise autonomous agents.

Computing Power Disruption: By leveraging computing resources, the platform offers higher token limits and lower latency.

Deep Governance: Enhanced system capabilities for orchestrating, controlling, and governing complex workflows and security in real-world business environments.

In the memo, D'Almeida expressed a nuanced view of the partnership with Microsoft, suggesting that over-reliance on Microsoft has limited OpenAI's ability to reach a broader range of customer scenarios.

AWS Bedrock: Since the collaboration began in late February, demand for this platform has been extremely high.

Stateful Runtime Environment: The memo highlights the advantages of Amazon's stateful runtime environment, which supports memory and context continuity across interactions, lowering the barrier for adoption in regulated industries.

Public Relations Campaign: Criticizing Anthropic's "Strategic Mistakes"

The memo includes pointed criticisms of Anthropic.

Computing Power Bottleneck: D'Almeida states that Anthropic's failure to secure sufficient computing power has led to customer throttling, which he calls a major "strategic mistake."

Narrow Positioning: He criticizes Anthropic for focusing too heavily on programming tools early on, which he argues will put it at a disadvantage in platform-based competition.

Questioning Revenue Figures: The memo accuses Anthropic of inflating its revenue run rate (around $30 billion) by counting payments to Amazon and Google as total revenue, which does not align with the net revenue recognition standards used by public companies.

The leaked memo suggests that by 2026, the focus of global AI competition has shifted from simply comparing model parameters to competing on enterprise implementation and ecosystem infrastructure.