Unverified50% confidenceReleaseExact time
Microsoft released the 4-billion-parameter hybrid computer-use agent MageVL Cowork 4B on Hugging Face, which uses a single compact policy model to decide autonomously when to click the interface and when to write code, achieving 66.0% on the Oathworld Verified benchmark and surpassing 27B- and 30B-parameter agents.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/1/2026
First Seen
Valid until: 1/30/2027
Sources
GPT-6.2 Leaked, Gemini 4 Spotted? AI Daily Roundup
bilibili一只AI风向标10/31/2026
Related Entities
Related Claims
UnverifiedMicrosoft has open-sourced Mage (called MajorVL Co-Work in the source material), a 4B-parameter desktop AI agent that scored 66% on the OSWorld Verified benchmark.74% similarUnverifiedV4-Pro uses a Mixture of Experts (MoE) architecture with a total parameter count of 1.6 trillion (1.6T) and 49 billion (49B) active parameters.69% similarUnverifiedDeepSeek released the first visual version weights of V4, with a model scale of 305B (approximately 305 billion parameters), along with a minimal inference implementation.66% similarUnverifiediFlytek open-sourced two small models of Spark X 2.5, the 4B and 1.7B, targeting edge-side agent scenarios, employing a hybrid attention architecture with native support for up to 1 million token context.65% similarUnverifiedThe CS4 launched by Cerebras is claimed to be the fastest AI inference hardware currently available, ultimately capable of running models with up to 100 trillion parameters at a speed of approximately 1,000 tokens per second.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1124465API
curl https://kongchang.com/api/v1/knowledge/claims/1124465MCP
get_claim(id=1124465)