Duan Yongping Buys Moutai; AI Agents, Quant Hires & Research

Trending News

Duan Yongping Buys More Moutai

On the morning of September 28, well-known investor Duan Yongping (Snowball ID: Da Dao Wu Xing Wo You Xing) posted on the Snowball platform, simply stating, "bought some Kweichow Moutai." The post indicates an average purchase price of approximately 1,230 RMB for 30,000 shares, totaling over 36.9 million RMB.¹

OpenAI Dev Day Opens Hours After ChatGPT Frontend Code Leak; Code-Named 'O' Agent Draws Attention

Recent leaks of ChatGPT’s frontend code and testing channels revealed a new always-on AI Agent, code-named "o." Designed for long-running autonomous tasks and complex workflows, this product is seen as OpenAI’s heavyweight weapon to directly compete with agent platforms like Meta Muse.

Dev Day begins at 1:00 AM Beijing Time on September 30.

Anthropic Releases Mid-Tier Model Sonnet 5.5, 30% Faster, Outperforms Opus 5.5 on Agent Coding Benchmarks

Compared to its predecessor, Sonnet 5, this model reduces token consumption rates by up to 30%, making it suitable for daily coding and office document tasks. Sonnet 5.5’s network capabilities are described as comparable to Opus 5, making it the first Sonnet version eligible for the same cybersecurity safeguards as Fable and Opus.

The API pricing remains identical to Sonnet 5, but due to token consumption optimization, the actual comprehensive call cost can be reduced by up to 30%. ⁴

Quantitative Research

GPT-6 Astra Processes Tax Workbooks Twice as Fast as GPT-5.6 Sol

In complex financial document processing, whether a model can balance speed with accurate understanding of user intent is key to its practical application. Basis tested GPT-6 Astra against GPT-5.6 Sol using a tax workbook with 50 tabs. The results showed that GPT-6 Astra completed the workbook twice as fast while demonstrating stronger user intent comprehension, giving Basis greater confidence in the model’s usability in real-world scenarios. ⁶

Phase Transitions and Market Impact Structures in Pure-Agent Order Books

What systemic dynamics emerge at the macro level when all participants in a limit order book are replaced by autonomous reinforcement learning traders? Does the regime switch of order flow occur smoothly or abruptly, and what structure do impact costs follow? The study formally embeds agent interactions in a simulated environment with micro-matching mechanisms, filling buy and sell sides entirely with reinforcement learning traders, and quantitatively characterizing order flow and impact costs tick by tick. The results show that pure-agent order books exhibit clear equilibrium phase-transition boundaries at regime switches, with market impact structures varying by regime, differing systematically from order books with human participation. This framework bridges agent simulation and phase-transition analysis, providing a testable quantitative starting point for understanding liquidity risk in purely algorithmic ecosystems. ⁷

Joint Optimal Transport Calibration of SPX-VIX Risk Scenarios

How can joint risk scenarios be generated from SPX and VIX market smiles without full recalibration? The study uses Guyon’s entropy-marginal optimal transport formula to jointly calibrate the two-market smiles and introduces perturbation methods to calculate the sensitivity of the calibrated coupling via Fisher information linearization. The resulting framework is model-agnostic, allowing risk scenarios to be generated directly from linearization results, bypassing full recalibration steps, and providing computable sensitivity measures for perturbation responses. ⁹

Big Tech Moves

Citadel Recruits from AI Labs, Quant Teams Continue to Expand

Citadel is extending its quantitative recruitment触角 into AI research institutions, recruiting researchers directly from AI labs to expand its quant teams. This continues the main trend of top hedge funds over the past two years: "quantizing AI talent." Strategy iteration increasingly relies on large models and machine learning backgrounds rather than traditional financial engineering. The signal for job seekers is clear—AI research experience commands higher pricing in quant institutions.

Millennium and Point72 Ramp Up Data Center Talent

Millennium and Point72, two multi-strategy giants, are adding expertise in data centers. This move comes amid renewed skepticism about the effectiveness of certain hedge fund strategies. Both firms are responding with infrastructure and data capabilities: computing power and data pipelines are becoming the new arms race for multi-strategy platforms. The domestic implication is that the competitive dimension for top institutions has shifted from "strategy" down to "computing power + data engineering."

Quant Career

BofA, Wells Fargo, JPMorgan 2027 Summer Internship Interviews Complete with No Responses; LA Roles Pending

A candidate, after three "Super Days," faces the dilemma of feeling good about their performance but receiving zero offers. Where is the problem?

The three Super Days on August 27, August 28, and September 17, spaced less than three weeks apart, correspond to Enterprise Credit, Commercial Banking, and C&SI business lines. This means preparation materials cannot be universal—Enterprise Credit focuses on financial statements and credit logic, Commercial Banking values client relationships and regional market understanding, and C&SI is closer to the intersection of markets and products. Their assessment focuses and linguistic frameworks differ.

The work location is the variable most easily overlooked after Super Days: approval rhythms, headcount availability, and offer distribution sequences often differ across offices. Candidates in non-core offices may misjudge their situation when seeing offers shared elsewhere, leading to incorrect decisions to wait or give up.

Three key takeaways can be drawn.

First, the end of Super Days does not mean the end of the process. Candidates should maintain their application and interview pace during the waiting period. Do not bet everything on one line; the opportunity cost of the hiring season far outweighs the fatigue of interviewing at multiple firms.

Second, information asymmetry can be partially offset—peer communities, mutual aid among similar candidates, and proactive posting to ask questions are low-cost intelligence-gathering methods, especially useful when target offices have no public hiring information.

Third, there is no strong correlation between subjective feelings and final results. "Feeling like I did well" does not guarantee an offer. What can be controlled is only the quality of preparation and the volume of applications.

The waiting period is also a review window: record every question asked, every stumbling block, and every reluctantly answered case, regardless of the outcome. These records will become preparation material for the next round. For those job hunting in non-core offices, incorporating geographic factors into expectation management in advance is more useful than post-hoc anxiety. ¹¹

Latest Quant Job Info: 21 Companies Open 626 New Positions

Top firm Jiukun Investment is hiring Quantitative Trading System Development Engineers/Experts.

Requirements include proficiency in C/C++, solid computer science fundamentals (data structures and algorithms, operating systems, computer networks), independent engineering judgment, ability to decompose and deliver problems under ambiguous requirements, proficiency in AI-assisted development, a bachelor’s degree in a computer-related field, and trading experience (preferred but not required).

Over the past three days, global quant giants have added 626 positions across 21 companies. Geographic distribution: 18 in the US (18%), 78 in the UK (76%), and 6 in China (6%), with the UK having the most positions.

The greatest common denominator for candidate requirements is a bachelor’s degree in a computer-related field, proficiency in C/C++ and computer fundamentals, ability to independently solve complex engineering problems, and use AI to accelerate delivery.

This indicates that the financial industry is partially accepting AI and vibe coding.

Want to get daily latest job postings and apply ahead of others? Please subscribe to our job delivery service via email. ¹⁴