Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Benchmark

AI overview

Sign in and the AI will write an overview from our coverage.

Headlines · 12

  1. Sierra in Talks to Raise $1.5B at $30B Valuation

    Enterprise AI customer-support startup Sierra, founded by former Salesforce co-CEO Bret Taylor, is discussing about $1.5 billion in new funding at a valuation of roughly $30 billion, according to two people familiar with the talks. That would nearly double the $15 billion-plus valuation from its funding round announced in May. Lightspeed Venture Partners, Greenoaks and Benchmark are in talks to invest, but the negotiations remain early and the terms could change. Sierra said in October that its annualized revenue had exceeded $400 million; the supplied reports do not independently verify that figure.

    The Information · 🔥 0
  2. Tendrils Compute raises $30M for parallel-first processor

    Cambridge-based startup Tendrils Compute announced a $30 million funding round led by Benchmark, with Visionaries Club, the UK Sovereign AI Fund, FR8, Antimatter, Cloudberry VC, Conviction, Embassy Ventures and SV Angels participating, according to cnBeta. Tendrils aims to challenge conventional CPU and GPU architectures with a general-purpose processor designed to scale to thousands of cores. Its approach uses interaction nets to run many small computations in parallel while reducing reliance on shared mutable memory and linear instruction sequences. The company is also developing the open-source Vine programming language for its parallel-first architecture. Tendrils has not yet published detailed physical-chip specifications or performance benchmarks, so the architecture's scalability and competitiveness remain unverified.

    cnBeta · 🔥 0
  3. Mercor raises $750M Series D at $20B valuation

    Mercor raised $750 million in a Series D funding round that valued the AI training and evaluation startup at $20 billion, double its valuation after the October 2025 Series C. Benchmark, General Catalyst, Nvidia, Salesforce, Zoom, Sapphire Ventures and others participated, while the round also included an employee tender offer whose allocation was not disclosed. Mercor said the proceeds will support model evaluation, post-training infrastructure, related research and environments for practicing enterprise workflows.

    The Next Web · 🔥 0
  4. Cowboy Space Unveils Big Iron Express Orbital Data Center Rocket

    Cowboy Space unveiled Big Iron Express, a purpose-built rocket designed to carry five racks of NVIDIA Vera Rubin GPUs into orbit, and plans to offer it as a launch service competing with SpaceX. A SpaceX launch also carried Google’s experimental satellite, which will send simple requests to Gemini while testing four TPUs in space, alongside Cowboy Space’s Reason-1 satellite. Starcloud is pursuing the broader orbital-compute market with Benchmark backing and a reported valuation of $2.3 billion.

    钛媒体 · 🔥 0
  5. Legora Raises $550M Series E at $8.8B Valuation

    Stockholm-based legal AI startup Legora raised $550 million in a Series E funding round led by Atreides Management, giving the company an $8.8 billion post-money valuation. Legora plans to use the proceeds to expand across North America, Europe and the Asia-Pacific region while continuing to develop its Legora aOS platform. The platform supports legal research, legislative tracking and AI-agent workflows. Legora says its annual recurring revenue has exceeded $200 million and that more than 140,000 legal professionals across over 2,400 law firms and in-house legal teams use its software in more than 80 markets.

    The Next Web · 🔥 0
  6. Personal-agent ranking: Instinct's $10B round and the entry-point bet

    A 潮涌AI ranking republished by 36Kr profiles 15 personal-agent startups outside the big platforms, led by Instinct, which the article says closed a $1B round at a $10B valuation on September 28 with Sequoia, Benchmark and Coatue participating — five months after being registered and after four rounds that took its seed valuation from $50M to $10B. The article says Instinct has no standalone app and no revenue, is still invite-only, and that invite links were listed on eBay for $200, citing Cybernews' verification of two listings on September 30. It also describes sector patterns: most of the 15 live inside SMS or iMessage, subscriptions are described as a starting point with transaction commissions as the intended end state, and Today AI founder 齐俊元 both calls the market "the next big thing" and concedes that product maturity and business value do not match current valuations and hype.

    36氪 人工智能 · 🔥 0
  7. Manus's survival paradox: shorter windows and token bills for agent apps

    36Kr republished an article from the WeChat account Xiutai analyzing Manus's "dependent innovation": the company does not train foundation models but calls external ones such as Anthropic's Claude and Alibaba's Qwen to build an agent that can open web pages, write code and deliver files. It launched in March 2025, passed $100 million in annualized recurring revenue eight months later, and in October 2026 raised more than $500 million at a valuation of about $4 billion, the article says. It argues that in the LLM era the lead time available to platform-dependent builders keeps shrinking, and that agents sit directly on model companies' main path, so heavier usage means bigger token bills — Ji Yichao's engineering notes put Manus's input-to-output token ratio at about 100 to 1. In October Manus released version 2.0 and Cue, which gives agents email accounts and digital wallets, an attempt to build user relationships that survive a change of underlying model.

    36氪 人工智能 · 🔥 0
  8. Instinct's rapid fade shows AI investors turning from hype to unit economics

    A 36Kr piece from the Tech商业 account looks at how quickly discussion of Instinct, the AI startup founded by 23-year-old Noah Shinn under the legal name Spear Street Technology, has faded. It reports that the company raised a Series B at a $2.5 billion valuation in late August and then a $1 billion Series C on Sept. 28 at a $10 billion valuation, with Sequoia, Benchmark and Coatue participating. The product is described as a "personal AI butler" with no standalone app and an invite-only beta, where users text or call an AI agent to handle email, flights and groceries; the team has 14 people, users just topped 100,000, and the founder says it is not charging users yet, while the article cites an annualized GMV of about $1 billion. The author argues the mood has shifted as Anthropic's expected valuation reportedly slid from about $2 trillion toward $1.5 trillion, Microsoft trimmed its 2026 calendar-year capex outlook from roughly $190 billion to about $175 billion, and Meta held guidance at $125–145 billion while cutting about 8,000 jobs.

    36氪 人工智能 · 🔥 0
  9. 23-year-old's 14-person AI agent startup Instinct hits $10B valuation

    Citing Reuters, 36Kr reports that Instinct, the personal AI assistant startup founded by 23-year-old Noah Shinn, announced a $1 billion funding round on September 28 with investors including Sequoia Capital, Benchmark and Coatue, at a reported valuation of about $10 billion — roughly 4x its $25 billion... valuation of $2.5 billion in late August — with a team of just 14 people. Rather than a standalone app, users reach the assistant by text or phone call to book flights, buy household goods, cancel subscriptions or contact merchants, and a group-chat agent feature has been opened to some users. Shinn has said annualized transaction volume on the platform is approaching $1 billion, more than half travel-related, though he did not explain the calculation and the figure is self-reported. The Instinct Selections recommendation feature launched in late September drew complaints after suggesting products such as luggage without a shopping request, leaving monetization, privacy and reliability still unproven.

    36氪 人工智能 · 🔥 0
  10. Amazon reportedly in talks to buy Decart for about $7B

    The Wall Street Journal reported that Amazon is negotiating to acquire San Francisco AI startup Decart in a deal that could be valued at around $7 billion, a report relayed by Bloomberg, IT之家 and others. Founded in 2023, Decart sells a cross-chip optimization layer called DOS that lets AI workloads move across Nvidia GPUs, Amazon Trainium and Google TPU, and also builds the real-time video model Lucy and the interactive-environment model Oasis. Reports say Anthropic discussed a roughly $6 billion acquisition in August and walked away after due diligence, while Nvidia was also named as a possible suitor; Amazon and Decart did not comment, and TechRepublic noted CTech floated a possible $5.5 billion valuation.

    The Next Web · 🔥 0
  11. Manus raises over $500M, reportedly at a $4B valuation

    On October 8, Manus parent Butterfly Effect announced a new funding round of more than $500 million, led by Boyu Capital and IDG Capital with existing backers Tencent, Sequoia China and ZhenFund participating; no valuation was disclosed, though Bloomberg previously reported a target of about $4 billion. It is Manus's first outside funding since it resumed independent operations on September 1, after Meta's acquisition was blocked by regulators in April and the founding team and investors bought the shares back at a $20 billion... correction: at a $20 billion figure was the buyback valuation reported by Bloomberg, with Tencent becoming the largest outside shareholder. On September 28 Manus released version 2.0 and a personal agent app, Cue; reports say its annualized revenue run rate has reached $400-500 million, though the revenue mix has not been disclosed.

    钛媒体 · 🔥 0
  12. Instinct Raises $1B Series C at Reported $10B Valuation

    Instinct announced a $1 billion Series C led by Sequoia, Benchmark and Coatue, reportedly lifting its valuation from $2.5 billion to $10 billion in one month. The 14-person personal-agent company operates without a mobile app, using text messages, calls and email to handle tasks such as travel bookings, price comparisons and orders, according to 36Kr and Datawhale. Founder Noah Shinn said about 40% of users bind a credit card after roughly three weeks and that retention among users connecting at least one sensitive asset is about 80%, though these figures were not independently audited.

    36氪 人工智能 · 🔥 0

Experience and discussion from the community

Share my Benchmark experienceAsk about Benchmark

cua: Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.

What problem it solves Computer-use projects need environments where agents can operate applications, run code, and interact with graphical interfaces within the same task. They also need repeatable ways to create tasks, evaluate agents, export trajectories, and generate data for training and research. Key features Shared desktop Spaces — Cua Spaces gives agents full desktops hosted on computers o…

github.comhttps://github.com/trycua/cua

heretic: Training-free language model modification.

What problem it solves Heretic addresses the difficulty of modifying language-model behavior to remove refusals, suppress slop, reduce sycophancy, or apply other behavioral changes. It provides a training-free approach that steers models through mathematical techniques rather than expensive retraining. Key features Configurable behavior changes — The default configuration suppresses refusals, whil…

github.comhttps://github.com/p-e-w/heretic

Switchyard: An open-source library that helps an AI agent choose which model handles each request.

What problem it solves AI agents need to choose among models while balancing task success, cost, and latency for their workloads. Applications also need flexibility in where routing runs and who controls model clients, credentials, retries, and service operation. Key features Multiple routing algorithms — Routing can be configured with Auto, Task, Execution, or Composite strategies, including mode…

github.comhttps://github.com/NVIDIA-NeMo/Switchyard

Vibe-Research: An investment research workbench for China A-shares, US stocks, and HK stocks.

What problem it solves The project brings market monitoring, watchlists, portfolio records, research, news, backtesting, and report management into one workbench across China A-shares, US stocks, and HK stocks. It also keeps research records and imported materials on the user's computer while allowing research to be conducted with local AI subscriptions or compatible model APIs. Key features Unifi…

github.comhttps://github.com/simonlin1212/Vibe-Research

Vibe-Trading: A personal trading agent that turns finance questions into research reports, runnable backtests and paper portfolios.

What problem it solves Finance research often requires combining market data, source documents, valuation methods, strategy testing and portfolio analysis across multiple workflows. This project provides an agent-based interface that turns those questions into research reports, backtests and paper portfolios, with optional supported live broker execution under user-defined limits. Key features Nat…

github.comhttps://github.com/HKUDS/Vibe-Trading

Hy-MT2: A family of fast-thinking multilingual translation models for complex real-world translation scenarios.

What problem it solves Real-world translation involves more than converting text between languages: tasks can require terminology, style, personalization, delimiter preservation, and structured-data constraints. Hy-MT2 is designed to support translation among 33 languages and follow translation instructions in multiple languages for complex real-world scenarios. Key features Instruction-driven tas…

github.comhttps://github.com/Tencent-Hunyuan/Hy-MT2

llmfit: A terminal tool that right-sizes open-source language models to the hardware and runtimes available on your system.

What problem it solves Selecting an open-source language model requires accounting for available memory, processor capabilities, GPU configuration, model size, quantization, and context requirements. The project addresses the difficulty of determining which models can run comfortably on a particular machine and comparing fit, speed, quality, and context before deployment. Key features Hardware aut…

github.comhttps://github.com/AlexsJones/llmfit

pocketpal-ai: A private AI assistant that runs entirely on your phone.

What problem it solves Many AI applications send messages to remote servers, where they may be logged and analyzed outside the user's view. This project addresses that concern by running language models, conversations, voices, and tools on the device without requiring an account or internet connection. Key features On-device language model chat — Run GGUF language models fully offline, including G…

github.comhttps://github.com/a-ghorbani/pocketpal-ai

BOXBOX: Evolving the Execution Framework During a Task

Performance optimization usually starts with a clear goal: completing the same work faster. What really takes time is identifying bottlenecks, designing experiments, comparing candidates, and then determining whether the differences come from the code or from measurement noise. After handing this process to an Agent, there is another kind of cost: repeatedly organizing logs, waiting for long-runni…

HarnessRSIBOXBOXPerfRaceAgent

Parallel Agent Teams: Small-Team Collaboration, Multi-Team Exploration

Optimizing a GPU kernel or completing a set of formal proofs often requires repeated trial and error: try a different implementation, run an evaluation, and keep iterating in whichever direction proves effective. When giving a model more test-time compute, the most direct approach is to let more agents try simultaneously; if the task only needs one best result, the value of the other attempts lies…

AgentPAT测试时计算多智能体Benchmark

arxiv.orghttps://arxiv.org/abs/2611.09798

memorilla: A memory system that gathers information from an entire document collection into query-dependent memory tokens for a frozen LLM.

What problem it solves A frozen LLM can only use information that fits in its context window, making it difficult to answer questions from large document collections or long interaction histories. Memorilla addresses this by reading a collection and supplying the LLM with a fixed number of query-dependent memory tokens. Key features Query-dependent memory compression — The system reads a complete …

github.comhttps://github.com/snap-stanford/memorilla

wealthfolio: An open-source, private portfolio tracker for investments, net worth, spending, and simulations.

What problem it solves Personal finance data often depends on accounts, cloud services, or optional integrations. Wealthfolio addresses this by keeping data on the user's device while supporting manual tracking, CSV import, portfolio analysis, and financial planning. Key features Portfolio and net worth tracking — Track investments across accounts and asset types, including options and short posit…

github.comhttps://github.com/wealthfolio/wealthfolio

ripgrep: A line-oriented search tool that recursively searches the current directory for a regex pattern.

What problem it solves ripgrep is a line-oriented search tool that recursively searches the current directory for a regex pattern. By default, it respects gitignore rules and automatically skips hidden files/directories and binary files. Key features Automatic ignore-aware filtering — By default the tool skips files ignored by .gitignore/.ignore/.rgignore files, hidden files and directories, and b…

github.comhttps://github.com/BurntSushi/ripgrep

watermarks-remover: A privacy-first agent skill and Python service for stripping AI provenance marks from text and files you own.

What problem it solves AI-generated content can contain provenance marks that remain in text, metadata, or media files even when the content is otherwise ready for use. The project addresses this by inspecting and removing marks across multiple vendor ecosystems and file formats while keeping the agent workflow separate from the processing service. Key features Multi-vendor text cleaning — Removes…

github.comhttps://github.com/guillaumemeyer/watermarks-remover

CodeAF: CodeAF is a coding harness built for open models, to get the most out of every dollar.

What problem it solves Managing multiple coding agents across separate terminals leaves users responsible for scheduling, checking, and merging their work. Teams also need a practical way to route tasks across models, automate recurring work, and operate the same development process in interactive or headless environments. Key features One workspace for every project — A single interface lists pro…

github.comhttps://github.com/Agent-Field/CodeAF

fleet: An open-source platform for IT and security teams to manage thousands of computers.

What problem it solves IT and security teams need one place to secure and maintain every computing device in an organization, across operating systems and cloud and container estates. Without that, device management, patching, software deployment, and compliance verification end up fragmented across separate tools. Key features Device management across every OS — Fleet manages Linux, macOS, Window…

github.comhttps://github.com/fleetdm/fleet

T3MP3ST: A multi-agent offensive-security framework, built to turn the AI coding agent you already run into a zero-day hunter.

What problem it solves Coordinated bug hunting is often gated by years of security practice, expensive tooling, and access to specialized workflows. The framework aims to make authorized reconnaissance, exploitation, and reporting more accessible while retaining explicit scope controls and reproducible evidence. Key features Agent-led security workflow — Operators follow a reconnaissance-to-exploi…

github.comhttps://github.com/elder-plinius/T3MP3ST

Agent Swarm: Organizational Growth and the Passing On of Experience

The Research system Anthropic unveiled in June 2025[1] organizes a research problem into a process of analysis, parallel search, and synthesis of results. Beyond model capabilities, how work is allocated and how results from different directions are connected have also become design questions that AI engineering needs to answer separately. Along this line, I want to ask a deeper question: if even …

EvoMapEvoXOpenClawSwarmGEP

evomap.aihttps://evomap.ai/

Why I Chose Personal Quantitative Investing

Quantitative hedge funds are becoming increasingly dominant. This is an undeniable fact. Leading institutions have data sources extending down to Level 2, tick-by-tick trades, and alternative data; computing power running backtests around the clock on thousands of machines; and talent pools packed with math and science PhDs from Tsinghua, Peking, Fudan, and SJTU. Individuals cannot compete on a si…

blog.hancic.sitehttps://blog.hancic.site/post/d0f10014-2ea7-4528-b5d4-0a188da05905