BlueRun Ventures Headlines | Moonshot AI's Intelligent Assistant Launches Beta for 2 Million Character Lossless Context Window

If we believe that lossless long-context capability is a foundational technology on the path to AGI, then a major breakthrough has arrived: Moonshot AI's Kimi intelligent assistant has achieved another leap in long-context window technology, expanding its lossless context length from 200,000 Chinese characters to 2 million characters.

BlueRun Ventures is an early investor in Moonshot AI. We look forward to the continued expansion of AI's possibilities — including complete code repository analysis and understanding, autonomous agents capable of completing multi-step complex tasks for humans, lifelong assistants that never forget critical information, and truly unified multimodal architectures, among others.

When Kimi first launched in October 2023, its approximately 200,000-character lossless context capability helped users unlock numerous new use cases, including professional academic paper translation and comprehension, legal analysis assistance, batch processing of dozens of invoices at once, and rapid understanding of API development documentation. It quickly gained strong user口碑 and rapid user growth.

Less than six months later, we are announcing today that Kimi has achieved another breakthrough in long-context window technology, with its lossless context length increasing by an order of magnitude to 2 million characters.

Starting today, Kimi with 2 million-character context support has launched "internal testing." Users with needs for ultra-long lossless context capabilities can apply for early access on the Kimi intelligent assistant web version at kimi.ai.

Going from 200,000 to 2 million characters — without taking the conventional gradual improvement approach — meant that the technical difficulty faced by the Moonshot AI team increased exponentially. To achieve better long-window lossless compression performance, their R&D and technical teams conducted native redesign and development across model pre-training, alignment, and inference, avoiding technical shortcuts like "sliding windows" and "downsampling" to overcome numerous fundamental technical challenges.

We believe that this order-of-magnitude increase in large model lossless context length will further help expand imaginations of AI application scenarios, including complete code repository analysis and understanding, autonomous agents capable of completing multi-step complex tasks for humans, lifelong assistants that never forget critical information, and truly unified multimodal architectures, among others.

To "throw out a brick to attract jade," here are a few example use cases for ultra-long lossless context:

After uploading hundreds of thousands of words of classic Texas Hold'em tutorials, users can have Kimi play the role of a poker expert to provide strategic guidance.

Upload a complete, nearly million-word traditional Chinese medicine diagnosis and treatment manual, and have Kimi provide diagnostic and treatment recommendations based on user questions.

Upload NVIDIA's complete financial reports from recent years, and have Kimi become an NVIDIA financial research expert to help users analyze and summarize important historical development milestones.

Upload source code from a code repository, and ask Kimi about any detail of the codebase — even legacy code with zero comments can help you quickly map out its structure.

Fields that once required 10,000 hours to master can now be approached at a junior expert level by Kimi in just 10 minutes. Users can discuss domain-specific questions with Kimi, use it to practice professional skills, or spark new ideas. With Kimi's support for 2 million-character lossless context, rapidly learning any new field becomes significantly easier.

Quickly organizing large volumes of materials is a common workplace challenge. Now Kimi can deeply read 500 or even more files in one go, helping users rapidly analyze all content and supporting natural language queries and filtering, dramatically improving information processing efficiency. For example, HR professionals can quickly have Kimi identify candidates with experience in a specific industry and computer science degrees from a recent batch of 500 resumes, enabling more efficient screening and candidate identification.

Uncovering subtle clues and mining deep details from full-length novels, stories, or screenplays is something many film and entertainment IP enthusiasts love to do. If you feed Kimi the complete Empresses in the Palace screenplay — hundreds of thousands of characters — and ask which details indicate that Zhen Huan's child belongs to the Prince of Guo, Kimi can dig deep into the emotional threads between Zhen Huan and the Prince across different time periods and scenes, uncovering the truth about their child, rivaling a "Zhen" scholar who has watched the series dozens of times.

"For AGI, lossless long context will be a very critical foundational technology. From word2vec to RNN, LSTM, and then to Transformer, essentially all historical architecture evolution has been about improving effective, lossless context length," Moonshot AI founder Zhilin Yang said in a previous interview. "Context length may have its own Moore's Law, but it only becomes meaningful scaling if you simultaneously optimize both length and lossless compression level."

Based on extensive user feedback, Kimi's 200,000-character lossless long context helped open up new worlds of AI applications and deliver greater value. But as users attempted more complex tasks and interpreted longer documents, they still encountered situations where conversation length exceeded limits. This is a direct reason why large model products need to continue expanding their lossless context length.

Additionally, Kimi's intelligent search capabilities are fundamentally enabled by its lossless long-context ability. Multiple sources that Kimi actively retrieves become part of the context passed to the model for reasoning. It is precisely because Kimi's large model supports sufficiently long context windows with sufficiently low information loss that the Kimi intelligent assistant can produce high-quality results and deliver a fundamentally different search experience.

Kimi can proactively search the internet based on user questions, analyze and synthesize the most relevant pages, and generate more direct, accurate answers. For example, users can have Kimi actively search for and compare the latest financial report data of two companies in the same sector, directly generating comparison tables and saving substantial research time. Traditional search engines typically can only return web links mixed with advertising based on user queries.

Another metric closely tied to large models' lossless context capability is instruction following. This manifests in two main aspects: first, whether the model can consistently follow user instructions and understand user needs across multi-turn conversations; second, whether the model can follow complex instructions that may run thousands or even tens of thousands of characters. Since launch, user feedback has shown that Kimi's multi-turn interaction and ultra-long instruction following capabilities are among its core product strengths.

With daily model capability upgrades and the launch of iOS, Android, Kimi intelligent assistant, and Web (kimi.ai) platforms, Kimi has become an indispensable AI assistant for more and more users in their work and lives. Following today's launch of 2 million-character ultra-long context internal testing applications, Moonshot AI will gradually open access to more users to experience Kimi's ultra-long lossless context capabilities, looking forward to co-creating intelligence with more users.

▶ More Content ◀

Originating in Silicon Valley, BlueRun Ventures was established in 2005 as a venture capital firm focused on early-stage startups.

Currently, BlueRun Ventures manages multiple USD and RMB dual-currency funds in China, with assets under management exceeding RMB 15 billion, making it one of the largest early-stage funds domestically. It invests primarily at Pre-A and Series A stages across technology, consumer, and healthcare sectors, having backed nearly 200 startups including Li Auto, Waterdrop, QingCloud, Guazi.com, Qudian, Ganji.com, Energy Monster, Gaussian Robotics, Songguo Mobility, Yuntu Semiconductor, Machenike, CloudSaints Intelligence, Anxin Netshield, and BioMap.

BlueRun Ventures has been ranked #1 on Zero2IPO's "China Top 30 Early-Stage Investment Institutions" and ChinaVenture's "China Best Early-Stage Venture Capital Firms TOP30," and was named among Preqin's Top 10 Global VC Fund Managers for Sustained High Returns.

The firm has also received consecutive honors from Forbes China, 36Kr, Cyzone, Caixin Media, CBNweekly, Jiemian, and other media organizations, including "China's Best Early-Stage Firm of the Year," "China Top Venture Capital Firm," "Most Founder-Friendly Early-Stage Firm of the Year," and "Most Influential Early-Stage Firm of the Year."