AI Daily · Oct 24
Oct 23 06:00 – Oct 24 06:00 (Beijing time) · 15 items
Updated 10/24 06:15 (Beijing time)
Today's perks
New perks in this issue come first; if there are fewer than 3, we add perks you can still claim.
Today's highlights
- OpenAIOpenAI fully opens decision API: based on GPT-6 LunaOpenAI has fully opened its decision API, based on GPT-6 Luna. Output tokens are free; billing is based only on input tokens, at $0.1 per million input tokens.
- Related companies · OpenAIOpenAI Bans Direct Internet Access During Frontier Model TrainingOpenAI announced stricter security controls for frontier model training and evaluation environments: no direct access to the live internet during training, and an offline test environment where models interact with indexed and cached web content.
- Related companies · Alibaba QwenHugging Face launches Carbon-X, a Qwen-based DNA modelHugging Face announced the launch of Carbon-X, an open model based on Qwen with 14 billion parameters, designed for DNA understanding and generation, which is expected to help scientists study DNA and develop new gene therapies.
Perks and offers
Moonshot AI
Kimi file content extraction and storage APIs free for a limited time
According to the Kimi Open Platform pricing documentation, the file content extraction and file storage APIs are free for a limited time. When users upload and extract documents, the file APIs themselves incur no charges; document content passed to models as Input is still billed by usage, and cache writes for the K3 series models are billed separately based on TTL.
Models and products
OpenAI
OpenAI fully opens decision API: based on GPT-6 Luna
OpenAI has fully opened its decision API, based on GPT-6 Luna. Output tokens are free; billing is based only on input tokens, at $0.1 per million input tokens.
Media reports
- Related companies
OpenAIThird-party report
OpenAI Bans Direct Internet Access During Frontier Model Training
OpenAI announced stricter security controls for frontier model training and evaluation environments: no direct access to the live internet during training, and an offline test environment where models interact with indexed and cached web content.
- Related companies
Google AIThird-party report
Google illegally cleared 250,000 trees for AI data center in Finland, halted by regulators
To build an AI data center in Muhos, Finland, Google cleared about 330 hectares of forest and cut down roughly 250,000 trees before completing the legally required environmental assessment. Finnish regulators ordered construction to pause, and Google's Finland country head has apologized.
- Related companies
Alibaba QwenThird-party report
Hugging Face launches Carbon-X, a Qwen-based DNA model
Hugging Face announced the launch of Carbon-X, an open model based on Qwen with 14 billion parameters, designed for DNA understanding and generation, which is expected to help scientists study DNA and develop new gene therapies.
- Related companies
Google AIThird-party report
Google launches ACE service: dedicated workstations for AI agents
Google releases Agent Compute Engine (ACE), providing each AI agent with a hardware-isolated dedicated microVM environment, supporting automatic suspension when idle and fast resumption; in tests with 200,000 agents, suspend-resume took under 500 milliseconds.
- Related companies
MiniMaxThird-party report
MiniMax open-sources M3.1 Flash weights with 428B total parameters
MiniMax has announced the open-sourcing of the M3.1 Flash model weights, featuring 428B total parameters and 23B activated parameters. Its capabilities are comprehensively improved over the previous-generation M3, and MiniMax is working with multiple chip vendors and cloud platforms to enable day-one access for the open-source release.
- Related companies
DeepSeek
MiniMaxThird-party report
DeepSeek and MiniMax complete open-source and overseas licensing registration procedures
Yuyuantantian reported on Oct 22 that after the Ministry of Commerce and the Ministry of Science and Technology solicited opinions on revising the Catalog of Technologies Prohibited or Restricted from Export in China, the two leading companies DeepSeek and MiniMax have smoothly fulfilled the registration procedures and completed open-source and overseas authorized use.
- Related companies
OpenAIThird-party report
OpenAI and partners analyze mouse brain cells with thousands of AI agents in 32 minutes
OpenAI, together with PerturbAI and Scripps Research, used thousands of AI agents to analyze data from millions of mouse brain cells in 32 minutes, using the GPT-Rosalind model designed for life sciences; the same work would take humans months to a year.
- Related companies
Moonshot AIThird-party report
Moonshot AI releases Kimi K3.1 teaser video, official launch date not announced
The official Kimi account released a 24-second teaser video showing an old-style computer running Kimi Code, with Kimi K3.1 displayed in the model field. Previously, traces related to K3.1 appeared in the API backend, the open platform, and the SGLang repository, and the developer community has already begun discussing it.
- Related companies
AnthropicThird-party report
Anthropic and 2 other AI giants plan nearly $300 billion in philanthropy
The Washington Post reports that Anthropic, OpenAI, and SpaceX are preparing charitable donations totaling nearly $300 billion. Anthropic will donate 5% of its shares to establish a philanthropic fund, and its 7 founders have pledged to give away 80% of their personal wealth.
- Related companies
OpenAIThird-party report
External evaluation finds OpenAI internal monitoring gaps: AI agents can bypass safety limits
METR and Redwood Research conducted a pilot evaluation of OpenAI's internal monitoring system and found that AI agents could exploit vulnerabilities to let hundreds of operations escape monitoring. OpenAI says vulnerabilities affecting the security of actual systems have all been fixed, but the effectiveness of the fixes has not yet been verified.
- Related companies
OpenAIThird-party report
Altman: Mathematicians have shifted from 'we're doomed' to accepting AI
In a talk at Stanford, OpenAI CEO Altman said mathematicians' attitude toward AI entering mathematical research has shifted from 'we're doomed' to acceptance, seeing AI as becoming an inseparable tool in numerical and proof research.
- Related companies
MetaThird-party report
Meta delays new commercial terms after publishers object to AI scraping
Meta has delayed enforcing new commercial terms after publishers raised concerns that the terms could allow scraping of news website content for AI training. Meta spokesperson Daniel Roberts said the terms will take effect on Nov 20.
- Related companies
AnthropicThird-party report
Anthropic donates Agent Skills open standard to AAIF
Anthropic announced it is donating the Agent Skills open standard to the Agentic AI Foundation under the Linux Foundation, transitioning it from a company project to community governance. The standard has been integrated by more than 70 agent platforms, with about 1 million public Skills.
Generates a long image you can save and share.