OS AI Field Notes | February 2026 Roundup

Throughout February, we covered key developments shaping the open-source AI landscape. This roundup consolidates the most consequential updates from the month.

OPENCLAW: FROM BREAKOUT TO BUYOUT

OpenClaw went from niche open-source agent to breakout phenomenon in a matter of weeks.

The project saw a record-breaking surge in GitHub stars, becoming one of the fastest-growing repositories in history as developers rushed to experiment with local-first autonomous agents.

That momentum was quickly tested. Security researchers reported info-stealer malware targeting OpenClaw configurations, exposing the risks of granting deep system access to open agents.

Days later, OpenAI acqui-hired the core team, signalling both validation and rapid consolidation at the enterprise layer.

ALIBABA QWEN

Alibaba continued expanding its open-source Qwen ecosystem. February saw the release of Qwen Image 2.0, advancing image generation and editing capabilities.

Earlier releases also included Qwen3-VL-Embedding and Qwen3-VL-Reranker, extending the multimodal stack.

ZHIPU AI AND MINIMAX

Zhipu AI released GLM-5, an open-source model with enhanced coding performance and agent task support.

The launch coincided with reports of a valuation increase amid a broader rally in Chinese AI equities.

MiniMax launched its updated M2.5 open-source model with expanded AI agent tooling. Its Hong Kong listed shares rose following the announcement.

Earlier in the cycle, MiniMax also launched MiniCPM-o4.5, a full-modal model supporting real-time interaction across vision, audio, and speech.

MOONSHOT AI

Moonshot AI released Kimi 2.5, an open-weight model approaching frontier proprietary benchmarks such as Claude Opus.

Reports indicate Moonshot is targeting a $12B valuation amid accelerating international demand.

Earlier, K2.5 variants gained strong benchmark traction and increased global visibility.

DEEPSEEK

DeepSeek published research on more efficient AI training methods, reflecting China’s focus on algorithmic efficiency amid hardware constraints.

NVIDIA

NVIDIA launched Alpamayo 1 on Hugging Face, an open-source collection of AI and vision-language models, simulation tools, and datasets aimed at autonomous vehicle reasoning.

NOUS RESEARCH

Nous Research released NousCoder-14B, trained in four days on 48 NVIDIA B200 GPUs. The model positions itself as an open-source competitor in coding-focused AI systems.

LIGHTRICKS

Lightricks open-sourced LTX-2, an audio-video generation model with open weights and training code optimized for consumer RTX GPUs.

TIINY AI

Tiiny AI announced a pocket-sized AI supercomputer capable of running 120B parameter open-source models locally, reducing reliance on cloud infrastructure.

OPENSCHOLAR

OpenScholar demonstrated performance advantages in literature reviews while maintaining citation accuracy. The tool can be freely tested or self-hosted.

PERPLEXITY

Perplexity AI rolled out a Deep Research upgrade and open-sourced the DRACO benchmark.

INDIA AI IMPACT SUMMIT AND SOVEREIGN FUNDING

India earmarked $1.1 billion for a state-backed venture capital fund focused on AI and advanced manufacturing.

At the same summit, Sarvam AI launched five new open-source models, including 30B and 105B parameter variants tailored to India’s linguistic diversity.

SINGAPORE AND UAE

Singapore committed over S$1 billion to AI research funding, strengthening regional open-model development.

The Institute of Foundational Model at Mohamed bin Zayed University of Artificial Intelligence launched K2 Think, a transparent sovereign model focused on advanced reasoning in math, science, and coding.

SOUTH KOREA

South Korea’s AI Basic Act entered its first full month of implementation. The 43-article framework establishes comprehensive national AI legislation, covering multiple regulatory domains including open-source systems.

YANN LECUN

Yann LeCun departed Meta and publicly argued that large language models are not the long-term path forward, advocating instead for world models and model-based planning through his new venture, Advanced Machine Intelligence.

SENTIENT FOUNDATION AND SENTIENT LABS

Sentient Foundation formally launched as a non-profit dedicated to ensuring artificial general intelligence remains open, decentralized, and aligned with humanity’s interests.

The Open AGI Summit returned to Denver, bringing together researchers and developers to explore decentralized intelligence and governance challenges.

Sentient Labs shipped an OpenClaw Payments Skill with a rules engine enabling programmable constraints for agent-based transactions.

Arena, Sentient’s benchmarking platform for enterprise AI agents, is launching March 4 with backing from Founders Fund, Pantera Capital, and Franklin Templeton, alongside ecosystem partners.

CLOSING REMARKS

February will likely be remembered for OpenClaw.

A project that went from relatively unknown to one of the fastest-growing open-source repositories on GitHub, pulled in massive developer attention, and was then acqui-hired by OpenAI within weeks. It was a clear signal that open-source agent infrastructure is moving quickly and attracting serious attention.

Alongside that, governments committed substantial capital to national AI strategies, and frontier open-weight models continued to close the performance gap with proprietary systems.