Stories

425 stories · updated Oct 28

Ongoing AI stories, each with every update in order and who confirmed it.

Microsoft Foundry Fine-Tuning

As of Oct 22, Microsoft Foundry Managed Fine-Tuning expands in public preview with four new open-source models and MAI-Code-1.1-Flash on Foundry resources.

LatestReported by MicrosoftMicrosoft Foundry Fine-Tuning adds open-source models, MAI-Code-1.1-Flash and multi-LoRA shared compute, per MicrosoftOct 22

Product1 update · last updated Oct 22

As of Oct 22, You.com launched Inference Router, an OpenAI-compatible API that routes across 12 open-weight models with a 99.9% uptime SLA, zero data retention, 100% US hosting, server-side web search and no platform fee, per MarkTechPost.

LatestReported by MarkTechPostYou.com launches Inference Router for 12 open-weight models with a 99.9% uptime SLA, per MarkTechPostOct 22

Company1 update · last updated Oct 22

Wuwai Zhiyu

As of Oct 22, 36Kr reports that children’s AI hardware firm Wuwai Zhiyu (物外智趣), founded by former Li Auto product staff, closed an angel round of tens of millions of yuan from Boreas Capital, angel investor Prof.

LatestReported by 36KrWuwai Zhiyu, a kids AI hardware startup from ex-Li Auto staff, raises tens of millions of yuan angel round, per 36KrOct 22

Company1 update · last updated Oct 22

Chain Reaction

As of Oct 22, 36Kr reports that Chain Reaction (链式反应), which aims to price AI security work on token usage, completed a multimillion-yuan seed round after coming out of the open-source community.

LatestReported by 36KrChain Reaction, an AI security startup using token billing, raises a multimillion-yuan seed round, per 36KrOct 22

Product1 update · last updated Oct 22

SpecEvo

On Oct 22, SpecEvo pairs cheap and frontier models for scientific discovery, cutting API cost by up to 80%.

LatestOfficialSpecEvo pairs cheap and frontier models for scientific discovery, cutting API cost by up to 80%Oct 22

Product1 update · last updated Oct 22

Agent4Sci

On Oct 22, Agent4Sci tests whether coding agents can beat published scientific ML models; Claude Opus improves them by 10.4%.

LatestOfficialAgent4Sci tests whether coding agents can beat published scientific ML models; Claude Opus improves them by 10.4%Oct 22

Product1 update · last updated Oct 22

Successive Scale Racing

On Oct 22, Successive Scale Racing keeps agent harness evolution improving as the task pool grows.

LatestOfficialSuccessive Scale Racing keeps agent harness evolution improving as the task pool growsOct 22

Product1 update · last updated Oct 22

SkillEval

On Oct 22, SkillEval builds per-skill ability profiles of LLMs from benchmark data.

LatestOfficialSkillEval builds per-skill ability profiles of LLMs from benchmark dataOct 22

Product1 update · last updated Oct 22

SANE

On Oct 22, SANE automatically generates scorer-specific tests for LLM judges.

LatestOfficialSANE automatically generates scorer-specific tests for LLM judgesOct 22

Product1 update · last updated Oct 22

LADDER

On Oct 22, LADDER pruning keeps LLM quality more robust at 70% sparsity and above.

LatestOfficialLADDER pruning keeps LLM quality more robust at 70% sparsity and aboveOct 22

Product1 update · last updated Oct 22

BRACE

On Oct 22, BRACE allocates RL rollouts for MCP tool-use agents by forecasting which tasks give mixed rewards.

LatestOfficialBRACE allocates RL rollouts for MCP tool-use agents by forecasting which tasks give mixed rewardsOct 22

Company2 updates · last updated Oct 22

DeepSeek funding

As of Oct 22, DeepSeek has raised $15 billion in one of China's largest funding rounds, Bloomberg reports, days after the US ordered federal agencies to remove its technology over national security concerns.

LatestReported by BloombergDeepSeek raises $15 billion days after US agencies were told to drop its technology, per BloombergOct 22

Company1 update · last updated Oct 22

China Telecom Xingchen

As of Oct 22, China Telecom AI Technology's Xingchen AGI lab released and open-sourced Xing4.0-29B-Intent with 2.75 million training samples, weights and an inference image, per QbitAI.

LatestReported by QbitAIChina Telecom's Xingchen lab open-sources Xing4.0-29B-Intent, a model for intent understanding, per QbitAIOct 22

Product2 updates · last updated Oct 22

Agent Vault

On Oct 15, Self Labs launched Agent Vault so AI agents can pay and sign in without ever seeing real card numbers or passwords.

LatestReported by The Verge1Password launches 1Password for Agents so AI agents can sign in without seeing passwords, per The VergeOct 22

Product1 update · last updated Oct 21

Claude Duet

On Oct 21, Anthropic opened a waitlist research preview of Claude Duet, a proactive agent app for Pro and Max subscribers in the US.

LatestOfficialAnthropic opens a waitlist research preview of Claude Duet, a proactive agent app for Pro and Max subscribers in the USOct 21

Product1 update · last updated Oct 21

Granite Speech

On Oct 21, IBM released Granite Speech 5.0 2B and a multilingual Granite Speech 5.0 Turbo CTC model.

LatestOfficialIBM releases Granite Speech 5.0 2B and a multilingual Granite Speech 5.0 Turbo CTC modelOct 21

Company9 updates · last updated Oct 21

Fireworks AI

On Oct 21, Fireworks serves DeepSeek V4.1 Flash 1017 at more than 1,500 tokens per second on its Ultrafast tier.

LatestOfficialFireworks serves DeepSeek V4.1 Flash 1017 at more than 1,500 tokens per second on its Ultrafast tierOct 21

Company8 updates · last updated Oct 21

Perplexity

On Oct 21, Perplexity introduced JOLT, a post-training method that matched GRPO on MATH with 6.5x fewer training tokens.

LatestOfficialPerplexity introduces JOLT, a post-training method that matched GRPO on MATH with 6.5x fewer training tokensOct 21

Product2 updates · last updated Oct 21

Visual Studio Code

On Oct 21, VS Code made the Copilot SDK-based harness the default for new sessions and added model comparisons.

LatestOfficialVS Code makes the Copilot SDK-based harness the default for new sessions and adds model comparisonsOct 21

Company2 updates · last updated Oct 21

US data-center buildout

As of Oct 21, the Verge reports that the rush of new and proposed data centers in Texas, and the gas power used to meet their electricity demand, could bring air pollution back to levels not seen since the 2000s.

LatestReported by The VergeNew data centers could push Texas air pollution back to 2000s levels, per The VergeOct 21

How we label

Official: An institution announces something that has already happened, through its own channel or a founder or executive account.
Statement by …: A named person's views, predictions, plans or comments, including remarks reported by media when the original is unavailable.
Reported by …: Only media or another third party is saying it. We name the outlet.
Unconfirmed: Circulating online (posts, comments, leaks). No company, named person or outlet has confirmed it.

Items older than 72 hours never go in Top stories and show their original date.