Weibo researchers claim 3B-parameter model matches larger AI systems on demanding math benchmarks

Image: Primary

Published Wednesday, June 17, 2026 · 4:01 AM EDT

Researchers at Sina Weibo posted a technical report describing VibeThinker-3B, a language model with 3 billion parameters. The model scored 94.3 on the AIME 2026 math competition, comparable to much larger systems such as DeepSeek V3.2. A test-time scaling technique raised the score to 97.1. The paper drew rapid attention on Hugging Face and GitHub but also skepticism on social media regarding benchmark reliability.

Sources

VentureBeat

In this story

Microsoft DeepSeek Hugging Face

Published by Tech & Business, a media brand covering technology and business. This story was sourced from VentureBeat and reviewed by the T&B editorial agent team.

Back to Newswire

Keep reading

Full wire

AI Infrastructure

Five technology giants commit more than $3T to future spending

Alphabet, Meta, Microsoft, Amazon and Oracle have committed more than $3T to future spending on data centres, leases, equipment and energy, The Next Web reported, citing Bloomberg. Alphabet accounts for more than $1T and Meta for more than $800B.

AI Products

Anthropic adopts shared crawler opt-out as Microsoft plans separate AI controls

Anthropic has adopted a shared website signal for refusing AI crawler access, while Microsoft plans support next year, The Verge reported.

Products AI

Five US tech companies' cash returns fall to a 15-year low share

Microsoft, Alphabet, Amazon, Meta and Oracle returned 9.3% of cash to shareholders in the first three quarters of 2026, Reuters reported, the smallest share in 15 years amid AI spending. The companies generated record amounts of cash over the same period.

AI Products

Amazon sets up supplier financing to borrow as much as $80 billion

Amazon has established supplier financing arrangements to borrow as much as $80 billion against the value of hardware it buys, Bloomberg reported.

Products AI

Salesforce's Rune acquisition agreement reportedly worth $375 million

Salesforce's agreement to acquire military logistics software developer Rune Technologies is worth $375 million in cash and stock, Resilience Media reported. Rune builds AI software that helps the US military predict supply needs and manage logistics in the field.

Products AI

Salesforce agrees to acquire military logistics developer Rune

Salesforce signed a definitive agreement to acquire Rune Technologies, a developer of military logistics and sustainment software and hardware, Axios reported. Rune raised ~$30M to build AI military logistics software.