| Mobile Web

Search results for Claude Sonnet 4.5

AI & Enterprise

Claude Code sends 33,000 tokens before processing prompts, rising to 75,000 in real use

Anthropic\'s AI coding tool Claude Code sends tens of thousands of tokens to an AI model before processing a user\'s instructions, a test found. AI consulting firm Systima compared Claude Code with open-source OpenCode and logged data relayed to the model. Claude Code sent about 32,800 tokens in a minimal setup versus about 6,900 for OpenCode. In a development environment with instruction files, MCP servers and plugins, Claude Code sent about 75,000 tokens upfront, Systima said.

AI & Enterprise

DeepSeek launches new V4 model preview versions

Chinese AI developer DeepSeek has released preview versions of its new V4 AI model, the Wall Street Journal reported. The company announced the launch on its official WeChat account. DeepSeek said V4-Pro has significantly improved agent coding capabilities and offers user experience ahead of Anthropic\'s Claude Sonnet 4.5. It said output quality nears Claude Opus 4.6 in non-thought mode but trails its thought mode. The model supports Huawei chips and features Sparse Attention technology. DeepSeek also launched the cheaper, faster V4-Flash.

AI & Enterprise

LG unveils EXAONE 4.5 multimodal AI model focused on stronger reasoning

LG AI Research on Wednesday unveiled EXAONE 4.5, a multimodal AI model that understands and reasons over text and images. The model is positioned as a preparatory step to expand modalities under the company’s proprietary AI foundation model project, K-EXAONE. LG AI Research said EXAONE 4.5 is strong at reading and reasoning over complex industrial documents and posted benchmark scores exceeding several rival models.