Skip to main content
AI Observernews
  • News
  • AI agencies
  • Tutorials
  • Interviews
  • Awards
Subscribe
  • Models and research
  • Agents and automation
  • Chips and hardware
  • Business and startups
  • Policy and regulation
  • Health and science
  • Robotics
  • Design and product
  1. Home
  2. News
  3. Models and research
  4. Language models

Language models

  • All
  • Benchmarks and evaluation
  • Image and video models
  • Language models
  • Open models
  • Research papers
Technician waxing a ski in a bright alpine workshop, a nod to the ski simulator Google says Gemini 4 Neon built

Language models

Gemini 4 Neon is faster than Carbon, and cheaper once December starts

Google's new middle model built a working chess game in every run of one outside test, roughly ties Carbon on an independent index and costs $3 per million input tokens.

Lars Strother · Nov 20, 2026 · 3 min read

Fabric speakerphone, two-button handheld toy, white home hub and bare circuit boards on a bright studio table

Claude Fable 5.2 designed three working gadgets and costs 20% less

Lars Strother · Nov 19, 2026

Man fitting curved paper panels into a lamp at a sunlit workshop table, like the lamp OpenAI says Astra helped design

GPT-6.2 Astra matches Fable 5.2’s price; an outside index ranks it second

Lars Strother · Nov 18, 2026

Developers working at desks in a bright open-plan office, the kind of team that pays per token for a model API

Alibaba opens Qwen4-Max at $1.50 per million input tokens, 1M context

Lars Strother · Nov 16, 2026

Gloved hands lowering an AI accelerator board into a server tray on a lab workbench, illustrating chips used for model training

Meituan says LongCat-2.5, trained on Chinese chips, nears Claude Opus 5

Lars Strother · Nov 2, 2026

More news

    Get the weekly email

    The week's most useful AI news, a tutorial and an interview.

    One email a week. Unsubscribe at any time.

    Most read

    1. Mistral releases Large 4 weights as outside tests put its coding second
    2. NVIDIA VFlash cuts robot plan latency about 80% on Jetson Thor
    3. What an AI agent is, how it differs from a chatbot and where it fails
    4. What the four big AI leaderboards measure, and why their rankings differ
    5. AWS lets AI agents run Step Functions workflows, but not rewrite them
    AI Observernews

    Independent news about AI and the people who build with it.

    We cover AI tools, agencies, tutorials and interviews for readers who use AI at work.

    About

    • About us
    • Contact us
    • Editorial policy
    • Corrections
    • Advertise with us

    Sections

    • News
    • AI agencies
    • Tutorials
    • Interviews
    • Awards

    Legal

    • Accessibility statement
    • Terms of use
    • Privacy Policy

    Back to top ↑

    © 2026 AI Observer. All rights reserved.

    • RSS feed