Chinese AI lab known for publishing detailed technical reports on mixture-of-experts LLM architectures.
- MiniMax Takes a Shot at OpenAI and Anthropic
- Meta Releases Its First Frontier AI Models
- Competing ARR Estimates & Fuzzy Valuation Math in the Spotlight as Markets Get Edgy
- South Korea Investigates Potential AI Agent Banking Attacks
- Kalshi's valuation is a bold prediction
- Anthropic IPO at Risk, Meta's Muse Pop, Token Prices Fall, Open Source Gains Share, Alignment Fails
- Anthropic's Co-Founders Push for Majority Control
- Kalshi's long odds come at a cost 🎲
- 🌄 The new frontier
- Satya Nadella on the AI Doomer Slowdown, Microsoft's Master Plan & Who Wins AI
- Open Models Change The Economics of AI
- AI researchers debate how close we are to recursive self-improvement
“DeepSeek has open-sourced DeepInfer, the inference stack behind its V4.1-Flash model, as Chinese AI labs keep pressing U.S. rivals on cost. DeepSeek charges $0.035 per million input tokens and $0.12 per million output tokens, versus $10 and $50 for OpenAI's GPT-6 Astra.”
“The Department of Homeland Security ordered federal civilian agencies to remove DeepSeek technology from their IT systems and stop buying its products. The ban also covers offerings from DeepSeek's affiliates and any third-party services built on DeepSeek.”
Source→“Winverse AI, a Chinese developer of AI for robotics, is in talks to raise 'tens of millions of dollars' in seed funding at a $400m valuation led by DeepSeek, per Bloomberg.”
“Winverse AI, a Chinese startup founded this year that is building AI models for robots, is reportedly raising tens of millions of dollars in a seed round led by DeepSeek at about a $400 million post-money valuation.”
Source→“Winverse AI, a startup focused on AI for robotics, is raising a seed round led by DeepSeek at a $400 million valuation, Bloomberg reported.”
“frontier capabilities defined by LLMs available from Chinese companies such as Zhipu (GLM) and Moonshot (Kimi) and DeepSeek.”
“DeepSeek is close to raising $12 billion in funding co-led by Tencent and battery company Contemporary Amperex Technology Co. per Bloomberg.”
Source→“All three startups were openly positioning their new models as the ones to pick if you're concerned about AI sovereignty.”
Source→“A hacker from China allegedly used AI tools from China-developed models, including DeepSeek, to steal data from tens of thousands of bank customers. The attacker used Claude Code to ask for help finding places to sell the stolen data, according to cybersecurity firm CrowdStrike.”
“Top Chinese models like GLM 5.3, Kimi K3, and DeepSeek V4.1 remain the most popular on the cloud platforms Baseten and Fireworks.”
Source→“DeepSeek... is reportedly in the market to raise at least $12 billion at a targeted valuation of roughly $74.6 billion. CATL and Tencent are among the company's largest investors.”
Source→“Open weight models are roughly 3 to 6 months behind the frontier, led by Kimi K3, DeepSeek and Qwen.”
“DeepSeek V4.1 Flash 'outputs very fast, but thinks for very long, so fast token output doesn't necessarily mean the task finishes faster'”
Source→“InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation”
Source→“InterEvolve more than doubles the success of the best calibrated program reaching 86.5% success vs. 34.6%”
Source→“DSH Desktop — Official app for DeepSeek’s open-source agent harness (172 votes)”
Source→“DeepSeek... has reached a $1 billion annualized revenue run rate, more than doubling in just a few months... The company is also targeting a $7.45 billion funding round at a roughly $74.5 billion valuation as it prepares for a potential Shanghai listing.”
Source→“DeepSeek is raising a 50 billion yuan ($7.5 billion) round at a valuation of 500 billion yuan ($75 billion) ahead of a planned IPO in Shanghai.”
Source→“That model climbed to No. 1 on OpenRouter's leaderboard with a 172% spike in usage this week.”
Source→“I feel like DeepSeek coming out was a fairly clear inflection point and signal”
Source→“DeepSeek this month released V4.1 Flash, which it says beats its previous flagship offering on a number of benchmarks with technical updates that allow the company to charge less for its intelligence.”
Source→“you could sense this person is very, very serious about believing in this thing”
Source→“DeepSeek beginning of 2025 with DeepSeek v3... [got a] massive decrease in KV cache size with multi-head latent attention... none of the major US companies... are definitely not using MLA”
Source→“DeepSeek — Cited for architectural choices... optimized around real-world inference efficiency”
Source→“are engaging in aggressive, malicious, and targeted distillation activities at an industrial scale.”
Source→“DeepSeek... reportedly raising a round at a $75 billion valuation, pursuing an IPO on Shanghai's STAR Market”
Source→“Around 80% of these pupils started using AI models such as Doubao and DeepSeek during that period.”
“There are also growing questions about the cost of AI inference, which has companies exploring cheaper models built by Chinese companies like Moonshot, DeepSeek, and Alibaba.”
Source→“DeepSeek, a Chinese frontier lab, is in talks to raise $7.4b at a $74b valuation.”
“DeepSeek credited with MLA as a major DeepSeek V2/V3 architectural contribution”
Source→“I discussed with Chen Jian, the DFlash author; he raised a point important to the whole model — even with DFlash's advanced draft model strategy, its relationship to pretraining is still very significant.”
Source→“DeepSeek is "about to close a funding round that values the Chinese AI startup at roughly $74 billion, with backers like CATL and some heavy-hitting new investors lining up, all while the company quietly gears up for a possible Shanghai IPO as soon as 2027."”
Source→“"DeepSeek certainly publishes really interesting work to compress that further and further. And they're making good progress."”
Source→“Systems like DeepSeek R1 deliver performance comparable to GPT-4 at close to 15% of the operating expense.”
“While DeepSeek's cheap, capable model triggered ~$1 trillion in market value erasure across AI-adjacent equities, Altman says it was a non-event internally at OpenAI's research level.”
“Actually owning the chip is going the wrong way long term. Short term, it makes sense for larger companies because they want to optimize chipsets for models... The opportunity is that level, which is from the loops to customization, multiple things, security.”
Source→“DeepSeek says its new experimental multimodal V4 Flash model nearly matches Anthropic's top Opus 4.8 model on multimodal agentic tasks”
Source→“DeepSeek says its new experimental multimodal V4 Flash model nearly matches Anthropic's top Opus 4.8 model on multimodal agentic tasks, as Chinese AI developers continue closing the performance gap with U.S. rivals.”
Source→“"The agent is 'Bob', running on OpenClaw with DeepSeek V4 as the primary model."”
Source→“The conventional framing is that the best AI agent = the best underlying model. Lael's experience inverts this. Switching from Gemini to DeepSeek to GPT barely registered as a disruption, while a broken embedding service effectively killed Bob's personality.”
Source→“The agent is 'Bob', running on OpenClaw with DeepSeek V4 as the primary model.”
Source→“The most joyful recent thing was when DeepSeek 0731 came out — I was extremely happy.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.