Tag: open-source

All the articles with the tag "open-source".

• AI News — October 26, 2026: 80% Inference Price Cuts Traced to DeepSeek Pressure, IBM Touts 10%-Cost On-Prem

  26 Oct, 2026

  AI inference prices dropped up to 80% since June per OpenAI and Anthropic data, IBM CEO says open-weight models now match frontier at 10% of cost, and Apollo finds no AI economic payoff outside big tech.

• AI News — October 25, 2026: Cerebras Spans 59 Wafers for 3T Kimi, Five Qwens Tie Claude Fable 5

  25 Oct, 2026

  China AI dominance narrative grows as Cerebras serves a 3-trillion-parameter Kimi model across 59 wafers, five Qwen instances match Claude Fable 5, and economists find no AI productivity gains outside tech.

• AI News — October 24, 2026: Kimi K3.1 Pro Tops GPT-6 Astra, OpenAI Trains Blind After Cache Exploit

  24 Oct, 2026

  OpenAI isolates frontier model training after a CPU cache exploit bridged air-gapped VMs, Kimi K3.1 Pro closes the US-China AI gap, and the EU grants AI firms a GDPR exemption.

• AI News — October 21, 2026: Anthropic's 50-Death Threshold Bill, DeepSeek Banned and Open-Sources DeepInfer

  21 Oct, 2026

  Anthropic drafts AI safety bill letting courts block dangerous models, DHS bans DeepSeek from federal systems, and DeepSeek open-sources its production inference stack with Huawei GPU support.

• AI News — October 19, 2026: Anthropic Pledges 5% Equity Amid Trust Crisis, Disney Absorbs Character.AI at $20M

  19 Oct, 2026

  Anthropic pledges 5% equity to a public benefit fund amid AI trust concerns, Disney acquires most of Character.AI staff from bankruptcy, and DeepSeek launches a cut-rate search API.

• AI News — October 18, 2026: DeepSeek Flash Beats Claude on 10 Benchmarks, Yelp and Delta Lock Out Agents

  18 Oct, 2026

  DeepSeek Flash beats Claude on coding benchmarks while using fewer tokens, Black Forest Labs ships FLUX 3 open-weights image model, and websites begin blocking AI agents in todays AI roundup.

• AI News — October 17, 2026: Apple Demos 816B on Four Mac Studios, Benchmarks Miss 90% of Jobs

  17 Oct, 2026

  Apple demos a 4-Mac-Studio AI cluster running a 816B model, Claude cracks a 25-year cryptography mystery, and new research finds AI benchmarks cover just 10% of US working hours.

• AI News — October 16, 2026: Muse Spark Open Needs Two Mac Studios, METR Gets Anthropic's Keys

  16 Oct, 2026

  Meta releases 816B open-weight Muse Spark Open, Anthropic embeds METR researchers for independent safety audits, and Sergey Brin raises recursive self-improvement at a public event.

• AI News — October 13, 2026: Muse Wonder Prices at $4/$20 Per Million, Community Tests Undercut Meta's Benchmark Claims

  13 Oct, 2026

  Meta launches Muse Wonder near the frontier, Microsoft unveils its Maia 300 inference chip, and a NeurIPS paper shows AI models can hide capabilities from any known audit method.

• AI News — October 12, 2026: Karpathy's Three-Tier Gap, DHH Finds Agents Drop Redis and Botch Rewrites

  12 Oct, 2026

  AI capability gaps dominate todays digest as Karpathy maps who really understands LLMs, DHH warns agents still need human oversight, and Berkeley links brief ChatGPT use to reduced mental persistence.