| Mobile Web

Search results for DeepSeek V4 Pro

AI & Enterprise

DeepSeek releases official V4 Pro model; 46 times cheaper than Claude with 5 percent performance gap

Chinese AI company DeepSeek has switched its flagship model to the official DeepSeek-V4-Pro-0813 build, without a separate announcement and with API prices unchanged. Existing performance assessments were based on a preview version, so the official build requires external verification. DeepSeek’s benchmarks show Anthropic’s Claude Fable 5 averaged 5.3 percent higher across nine comparable agent benchmarks, while DeepSeek is far cheaper on mixed input-output pricing.

AI & Enterprise

Why Nvidia wants to develop even large AI models itself

Nvidia is speeding up its push to develop and distribute its own open-source AI models as part of an effort to expand hardware demand, although some see a risk of competition with AI companies that buy its chips. The Information reported Nvidia is developing the open-source large language model Nemotron 4 with at least 1 trillion parameters. Nvidia has also rolled out lightweight models and free routing software and increased spending on computing resources through cloud contracts.

AI & Enterprise

Local PC LLMs gain traction as tech firms step up competition

Competition among tech companies is intensifying over lightweight AI models as costs rise and more firms turn to open-source options for basic tasks. Local models that run on personal PCs are gaining prominence. DeepSeek released its V4 Flash coding model and reports say it shows strong performance at low API prices. Meta plans its Muse Glimmer laptop model family, highlighting on-device privacy. Nvidia introduced its open-source Nemotron 3.5 Lightning for PC GPUs.