# AIVIO News

> AIVIO News tracks artificial intelligence daily: model releases, research, funding, policy and the leaderboards that say which system is actually ahead.

AIVIO News is published by AIVIO Technologies d.o.o. (https://aiviotechnologies.com), an AI engineering company in Slovenia. Articles are written by a language model from the public sources each one links, with no human editor in the loop; the method is described in full at https://aivionews.com/about/. Updated every day at 07:00 CET.

## How to cite this site

- Each article lists its own sources at the bottom of the page. Prefer citing those originals.
- Leaderboard numbers are copied unmodified from the operator named on each board; cite the operator, and this site only for the snapshot date.
- Canonical URLs never change. Full machine-readable index: https://aivionews.com/api/articles.json

## Leaderboards, current leaders

- [Best overall chat model](https://aivionews.com/leaderboards/chat/): Claude Opus 5 Max (Anthropic), score 1505. Source LMArena, published 2026-09-01.
- [Best model for coding](https://aivionews.com/leaderboards/coding/): qwen3.8-max-0902 (Alibaba), score 1691. Source LMArena, published 2026-09-01.
- [Best agent model](https://aivionews.com/leaderboards/agents/): Claude Opus 5 (High) (Anthropic), score 13.8. Source LMArena, published 2026-08-31.
- [Best model for images and documents](https://aivionews.com/leaderboards/vision/): Claude Fable 5 (Anthropic), score 1329. Source LMArena, published 2026-08-27.
- [Best model for search](https://aivionews.com/leaderboards/search/): GPT 5.6 Sol Xhigh (OpenAI), score 1257. Source LMArena, published 2026-08-24.
- [Best image generator](https://aivionews.com/leaderboards/image-generation/): gpt-image-2 (medium) (OpenAI), score 1380. Source LMArena, published 2026-08-07.
- [Best video generator](https://aivionews.com/leaderboards/video-generation/): Gemini Omni 1.1 Flash (Google), score 1515. Source LMArena, published 2026-08-27.
- [Cheapest capable models](https://aivionews.com/leaderboards/price/): Mistral Nemo (Mistral), $0.022/1M tokens blended. Source OpenRouter, published 2026-09-02.

## Sections

- [Models](https://aivionews.com/models/): Every new frontier model, open-weights drop and capability claim, with the benchmark numbers that back it or fail to.
- [Research](https://aivionews.com/research/): Papers worth the time: new architectures, training methods, interpretability results and the negative results nobody else reports.
- [Business](https://aivionews.com/business/): Who is paying for artificial intelligence, who is making money from it, and what the numbers behind the announcements actually say.
- [Policy](https://aivionews.com/policy/): The EU AI Act, US executive action, copyright litigation and safety commitments — what changed, and what it obliges anyone to do.
- [Tools](https://aivionews.com/tools/): Agents, coding assistants, APIs and the products built on top of them, judged on whether they work rather than on the launch video.
- [Hardware](https://aivionews.com/hardware/): GPUs, accelerators, interconnect, data-centre build-outs and the power bills that increasingly decide who can train what.

## Recent articles

- [Trump administration backs OpenAI in copyright fight with The New York Times](https://aivionews.com/policy/trump-administration-backs-openai-in-copyright-fight-with-the-new-york/) — 2 Sept 2026. The Trump administration filed a 20-page amicus brief backing OpenAI's fair use defense, arguing that limiting AI training would harm U.S. competitiveness in artificial intelligence.
- [Anthropic drops Claude Fable 5.1 with cheaper pricing and lighter restrictions](https://aivionews.com/models/anthropic-drops-claude-fable-5-1-with-cheaper-pricing-and-lighter/) — 2 Sept 2026. Claude Fable 5.1 costs up to 45 percent less for complex agentic tasks and blocks fewer basic biology questions than its predecessor, Anthropic says.
- [OpenAI's Astra model raises alarm with harder-to-monitor reasoning](https://aivionews.com/research/openai-s-astra-model-raises-alarm-with-harder-to-monitor-reasoning/) — 2 Sept 2026. OpenAI's Astra uses a looping technique called recurrent depth that processes queries multiple times, and safety researchers warn wider use could make AI reasoning impossible to monitor.
- [Palo Alto Networks buys IT automation startup Console for $500 million](https://aivionews.com/business/palo-alto-networks-buys-it-automation-startup-console-for-500-million/) — 2 Sept 2026. The $500 million cash-and-stock deal folds Console's AI agents into Palo Alto Networks' Cortex platform, marking the cybersecurity giant's seventh acquisition of 2026.
- [Google DeepMind launches Gemini 3.8 Flash and a cybersecurity version](https://aivionews.com/models/google-deepmind-launches-gemini-3-8-flash-and-a-cybersecurity-version/) — 2 Sept 2026. The new model pushes deeper multi-step reasoning and tool use for coding agents, while the cyber-focused variant is limited to vetted defenders under a new access program called Fairwind.
- [OpenAI says Astra is first model to cross Critical cybersecurity threshold](https://aivionews.com/policy/openai-says-astra-is-first-model-to-cross-critical-cybersecurity/) — 2 Sept 2026. The classification, made under OpenAI's internal Preparedness Framework, triggers stronger safeguards before release, though the company has not detailed what those safeguards involve.

## Optional

- [Full text of recent articles](https://aivionews.com/llms-full.txt)
- [RSS feed](https://aivionews.com/rss.xml)
- [Archive](https://aivionews.com/archive/)
- [How this site is made](https://aivionews.com/about/)