Google is testing an unreleased Gemini 4 checkpoint codenamed Carbon on its internal Jetski platform. Leaks report coding capability rivaling Claude Opus 5.5.
Benchmarks & Hype Checks
89 Articles in this Category
Independent evaluations of synthetic benchmark claims: ARC-AGI-3, SWE-bench Verified, HumanEval, and live latency audits.
Paradigma reports 94.01% on AIME 2026 for its 1B math model. Off-scope failures and specific prompt and runtime requirements define its practical use.
Cloudflare Clef scores typed choices from text and images. The 27B model costs $0.24 per million input tokens; hosted image limits and latency need a close look.
Microsoft reports 61.5% on SWE-bench Verified for FrogNano 4B. A five-tool agent interface accounts for a large gain, with cost and licensing caveats.
Ling 3.1 Flash is available through hosted APIs, with 560 billion total parameters and about 25 billion…
Microsoft has released MAI-Transcribe-2-Streaming for live speech recognition and MAI-Voice-2.1, including a faster Flash variant, for speech…
onPanda is an open-source annotation interface that lets a human replace the first wrong token in a…
Kyutai’s Voice of Reason takes a spoken math question and generates an answer through a speech-native model…
Researchers tested 21 vehicles and 30 companion apps. Third-party contacts and exposed app identifiers are distinct findings with different evidence.
Backblaze reports 1.73% annualized failures for Q2 2026. Drive-days, fleet age and sample size explain why that figure is not a home-NAS forecast.