First Move  ·  AI Software  · 
Inside the labs — Monday morning, 20 July

Moonshot AI Unveils Kimi K3, Challenging Frontier Models; Anthropic, OpenAI Move Towards IPOs

New open-weights models are pushing performance boundaries, while major labs prepare for public markets amidst shifting financing structures and product adjustments.

ShareWhatsAppXLinkedIn

The story

The AI software landscape saw significant movement this week, with China's Moonshot AI making a notable entry by releasing Kimi K3, a multimodal open-weights AI model. This new model demonstrates frontier-level performance, particularly in coding, agentic knowledge work, and autonomous execution, featuring 2.8 trillion total parameters and a one-million-token context window.

Kimi K3's benchmark results place it close to, and in some cases surpassing, established models like Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol, especially in web development tasks. This release signals a continued global push in AI capabilities, with open-weight models increasingly competing with proprietary offerings.

Concurrently, the financial gears of the industry are turning, with Anthropic reportedly meeting bankers ahead of a potential IPO later this year, and OpenAI also consolidating power in anticipation of its own public offering. These financial maneuvers, coupled with warnings about the AI infrastructure boom shifting from cash-flow to debt funding, underscore a maturing industry balancing rapid innovation with growing capital market scrutiny.

Who moved

Moonshot AI

What Changed: The lab released Kimi K3, a multimodal open-weights AI model with 2.8 trillion total parameters.

Consequence: Kimi K3 exhibits frontier-level performance in coding and agentic knowledge work, challenging existing models from other labs.

Anthropic

What Changed: Anthropic adjusted usage caps for Claude Fable 5 on its Max and Team Premium plans, cutting them by 50%.

Consequence: This pricing shift is set to affect developers relying on Claude Fable 5, pushing Pro users towards API pricing.

OpenAI

What Changed: Greg Brockman consolidated more power within the company.

Consequence: This move is explicitly framed as occurring ahead of OpenAI's anticipated IPO.

Microsoft

What Changed: Microsoft launched Project Perception, a multi-model AI security tool.

Consequence: The tool leverages models from Anthropic, OpenAI, and its own, positioning itself as a more affordable alternative to Mythos 5 with broader distribution advantages.

Tom Blomfield

What Changed: Blomfield departed Y Combinator to join Anthropic's compute team.

Consequence: This high-profile hire signals Anthropic's continued focus on strengthening its underlying compute capabilities.

New models

Kimi K3

Lab: Moonshot AI

What: A multimodal open-weights AI model with 2.8 trillion total parameters and a one-million-token context window.

Use: It offers frontier-level performance for coding, agentic knowledge work, and autonomous execution.

Codeex Micro

Lab: OpenAI (in collaboration with WorkLoud)

What: A compact joystick designed to streamline interaction with AI agents.

Use: This innovative controller aims to make AI development more intuitive and reduce friction for teams creating AI-driven applications.

Market signals

Anthropic is reportedly meeting with bankers and investors ahead of a possible IPO later this year.

Implication: This indicates a significant step towards public markets for one of the major AI labs, aligning with broader industry trends towards IPOs.

The Bank for International Settlements (BIS) published a bulletin warning that the AI infrastructure boom is shifting from cash-flow funding to debt, with private credit playing a growing role.

Implication: This suggests increasing financial risk in the AI infrastructure sector, moving it from a purely technology story to a systemic capital-markets concern.

Fireworks AI reached a $17.5 billion valuation.

Implication: This highlights strong investor confidence and significant capital flow into enterprise AI tools that specialize general-purpose models.

Alphabet's stock fell following reports of a delay for Google DeepMind's Gemini 3.5 Pro.

Implication: This demonstrates how even minor delays in key model releases can impact market perception and investor confidence in major tech companies.

What we'll be watching

Reporting + analyst voices: grounded via Google Search at publish time.