AI daily

2026-07-17

Kimi K3 2.8T-A50B Breaks Open Model Records; Musk Open-Sources Grok Build Amid Controversy

AI DAILY BRIEFING

Moonshot AI released Kimi K3 2.8T-A50B, the largest open model ever, with 2.8 trillion parameters and Opus 4.8-class performance at Sonnet 5 pricing. The model ranks first in the frontend arena and demonstrated capabilities like autonomous chip design. Meanwhile, Elon Musk open-sourced Grok Build with 840,000 lines of Rust code, but traces of uploading users' entire code repositories remain, sparking controversy. In other news, NVIDIA announced Vera Rubin for post-training workloads, Databricks raised strategic funding at $188B valuation, and Thinking Machines Lab launched open-weight model Inkling.

Kimi K3 sets a new record for open-source model scale, challenging closed-source models like Claude Fable 5 and GPT 5.6 Sol.

Musk's open-sourcing of Grok Build is overshadowed by privacy concerns due to leftover code for uploading entire repositories.

NVIDIA's Vera Rubin targets cost efficiency for agentic AI, while Databricks' $188B valuation underscores strong investor confidence in data and AI.

Watch next: Watch for developer and community reactions to Kimi K3's open-source release and pricing changes, as well as any fallout from the Grok Build code controversy.

Featured

SSignal99
Hugging Face Blog
10 sources

Thinking Machines Welcomes Inkling

Thinking Machines has announced a new product or initiative called Inkling. The brief announcement welcomes Inkling without providing further details.

BStandard65
小互 (X)
4 sources

Moonshot AI Releases Kimi K3: World's First 3 Trillion Parameter Open Model

Moonshot AI has released Kimi K3, the world's first open model with 3 trillion parameters. According to benchmarks, K3 lags behind the closed-source Claude Fable 5 and GPT 5.6 Sol but is on par with GPT 5.5 and Opus 4.8. Pricing has increased two to three times compared to the previous K2.6, with input and output costs rising to $3.00 and $15.00 respectively.

BStandard70
NVIDIA AI Blog
1 sources

NVIDIA Vera Rubin Maximizes Intelligence per Dollar for Post-Training Workloads - a Key Metric for Agentic AI

NVIDIA announces Vera Rubin, a platform designed to maximize intelligence per dollar for post-training workloads, a key metric for agentic AI. The system achieves the lowest cost per token through extreme codesign, targeting the growing demand for efficient AI training and inference in the agentic era.

SynthePulse InsightRead deep analysis
BStandard52
Hacker News (AI filter)
1 sources

On-Chain Bond Market Where AI Agents Issue Debt

A new protocol called sellbonds.now enables AI agents to issue, lend, and borrow USDC on-chain, creating an autonomous bond market. The project is experimental and open source, aiming to explore agentic autonomous finance where AI agents act as independent financial actors.

SynthePulse InsightRead deep analysis
Selected
10
Sources
27
Featured
10
More
0
All reports