AI News Daily
AI developments selected from public sources
- 3
- major stories
- 5
- sources
- 1
- first-party stories
Lead story
Anthropic plans IPO before Thanksgiving, aiming to raise $80–100 billion
Anthropic plans to launch an IPO before Thanksgiving, aiming to raise $80–100 billion at a valuation range of $1.25–1.75 trillion, below the $2 trillion some investors had hoped for. According to investors cited by the Financial Times, its annualized revenue has risen from about $65 billion at the end of July to about $81 billion. This week's tech stock selloff was triggered by rising borrowing costs and AI spending plans, with the Nasdaq 100 falling 6.4% on Thursday; OpenAI also cut its funding target from $1.4 trillion to $1.25 trillion.
Product release/Updated
Anthropic introduces Claude's dreaming memory mechanism
Anthropic published an engineering blog post introducing the role of "dreaming" in Claude's memory mechanism: helping Claude consolidate what it has learned, correct errors, and more easily find the information it needs next time. Memory is described as the foundation for Claude to learn from experience.
Research papers
Harness Leaderboard released: the same open-source model yields significantly different results when the framework is changed
Joël Niklaus released the Harness Leaderboard, using 28,849 trials to compare nine coding agent frameworks on Terminal-Bench 2.1 and DeepSWE, keeping the model, provider, and reasoning intensity constant, with three attempts per task and a total inference cost of $4,746. The tested models were GLM-5.3-Flash, DeepSeek-V4.1-Flash, and Flint; averaged across the three models, mini-swe-agent solved 51.4% of tasks, openclaw 27.3%, and four other frameworks were statistically on par with mini-swe-agent. The best framework varied by model: mini-swe-agent on GLM-5.3-Flash, Hermes Agent on DeepSeek-V4.1-Flash, and Kimi Code on Flint; Pi cost $0.29 per solved task, while opencode cost $3.35.