Liquid AI’s LFM2.5-24B-A2B is an open-weight 24B MoE reasoning model with 2B active parameters for fast on-device agents.

Auto research — the practice of letting an AI agent loop through trial and error on a codebase to optimize a single score — is the biggest ...

Claude Haiku 5.5 launches with 90% lower input pricing and a 72.4% score on OSWorld 2.1 - matching the human baseline on the leading AI computer-use benchmark, the first small model to reach that ...

Microsoft's experimental Windows ML update adds a path for running GGUF models. Official documentation and packages show no NPU support, limited generation controls, different distribution ...

Researchers have unveiled byteification, a two-stage conversion method that retrofits existing subword-based large language models into byte-level models matching their performance at under one ...

Claude Haiku 5.5 launched with lower API rates and a 1M-token context window. See both price tiers and what Anthropic’s 75% ...

Haiku 5.5 makes repetitive AI work cheaper, but marketing teams need to check prompt bands, quality and total task cost.

A new paper from Zhuowen Tu's group proposes replacing the fixed 256-token grid used in autoregressive image models with a hierarchical quadtree.

Why does AI cost more in Hindi and Tamil? Discover the tokenizer tax and the hidden challenges of building multilingual AI ...

Claude Haiku 5.5 brings 1M context, adjustable effort, and $0.10 input pricing to high-volume subagent and browser workloads.

Claude Opus 5 uses nearly 3x more tokens for Russian than English. Here's why AI tokenization makes some languages more ...

OKXICE’s notice under the SEC’s new innovation exemption lists Nvidia, Tesla, Apple and 60 other stock tokens, to trade against stablecoins in Uniswap pools on OKX’s X Layer.