Qodo Aware Deep Research achieves 80% accuracy on new coding benchmark, surpassing OpenAI’s Codex at 74%, Anthropic’s Claude Code at 64%, and Google’s Gemini CLI at 45%

Qodo, the agentic code quality platform, announced Qodo Aware, a new flagship product in its enterprise platform that brings agentic understanding and context engineering to large codebases. It features the industry’s first deep research agent designed specifically for navigating enterprise-scale codebases. In benchmark testing, Qodo Aware’s deep research agent demonstrated superior accuracy and speed compared to leading AI coding agents when answering questions that require context from multiple repositories.

AI has made generating code easy, but ensuring quality at scale is now even harder. Modern software systems span hundreds or thousands of interconnected code repositories, making it nearly impossible for developers to maintain a comprehensive understanding of their organization’s entire codebase. While current AI coding tools excel at single-repository tasks, they cannot traverse the complex web of dependencies and relationships: the 2025 State of AI Code Quality report found that more than 60% of developers say AI coding tools miss relevant context. Qodo Aware addresses this limitation with a context engine that powers deep research agents that can automatically navigate across repository boundaries.

“Developers don’t typically work in isolation, they need to understand how changes in one service affect systems across their entire organization and how those systems evolved to their current state,” said Itamar Friedman, co-founder and CEO of Qodo. “Our deep research agent can analyze impact, dependencies and historical context across thousands of files and hundreds of repositories in seconds, something that could take a principal engineer hours or days to trace manually. This eliminates the traditional speed-quality tradeoff that enterprises face when adopting AI for development, while adding the crucial dimension of understanding not just what the code does, but why it was built that way.”

Also Read: AiThority Interview with Tim Morrs, CEO at SpeakUp

Qodo Aware features three distinct modes, each powered by specialized agents for different use cases. The Deep Research agent performs comprehensive multi-step analysis across repositories, making it ideal for complex architectural questions and system-wide tasks. For quicker code Q&As, the Ask agent provides rapid responses through agentic context retrieval, and the Issue Finder agent searches across repos for bugs, code duplication, security risks, and other hidden issues. These agents can be used to get direct answers, or integrated into existing coding agents, like Cursor and Claude Code, as a powerful context retrieval layer, enhancing their ability to understand large-scale codebases.

Qodo Aware uses a sophisticated indexing and context retrieval approach that combines Language Server Protocol (LSP) analysis, knowledge graphs, and vector embeddings to create deep semantic understanding of code relationships. For enterprises, this means developers can safely modify complex systems without fear of breaking unknown dependencies, reducing deployment risks and accelerating release cycles. Teams report cutting investigation time for complex issues from days to minutes, even when working across massive, interconnected codebases with more than 100M lines of code.

Along with these capabilities, Qodo is releasing a new multi-repository dataset for evaluating coding deep research agents. The dataset includes real-world questions that require information that spans multiple open source code repositories to correctly answer. On the new DeepCodeBench benchmark, Qodo Aware achieved 80% accuracy, while OpenAI Codex scored 74%, Claude Code reached 64%, and Gemini CLI correctly solved 45%. Importantly, Qodo Aware Deep Research took less than half the time of Codex to answer, enabling faster iteration cycles for developers.

Qodo Aware has been integrated directly into existing Qodo development tools – including Qodo Gen IDE agent, Qodo Command CLI agent, and Qodo Merge code review agent – bringing context to workflows across the entire software development lifecycle.. It is also available as a standalone product accessible via Model Context Protocol (MCP) and API, enabling integration with any AI assistant or coding agent. Qodo Aware can be deployed within enterprise single-tenant environments, ensuring code never leaves organizational boundaries, while maintaining the governance and compliance standards enterprises require. It supports GitHub, GitLab, and Bitbucket, with all indexing and processing occurring within customer-controlled infrastructure.

Alexander Mordvintsev showed me two clumps of pixels on his screen. They pulsed, grew and blossomed into monarch butterflies. As the two butterflies grew, they smashed into each other, and one got the worst of it; its wing withered away. But just as it seemed like a goner, the mutilated butterfly did a kind of backflip and grew a new wing like a salamander regrowing a lost leg.

Mordvintsev, a research scientist at Google Research in Zurich, had not deliberately bred his virtual butterflies to regenerate lost body parts; it happened spontaneously. That was his first inkling, he said, that he was onto something. His project built on a decades-old tradition of creating cellular automata: miniature, chessboard-like computational worlds governed by bare-bones rules. The most famous, the Game of Life, first popularized in 1970, has captivated generations of computer scientists, biologists and physicists, who see it as a metaphor for how a few basic laws of physics can give rise to the vast diversity of the natural world.

In 2020, Mordvintsev brought this into the era of deep learning by creating neural cellular automata, or NCAs. Instead of starting with rules and applying them to see what happened, his approach started with a desired pattern and figured out what simple rules would produce it. “I wanted to reverse this process: to say that here is my objective,” he said. With this inversion, he has made it possible to do “complexity engineering,” as the physicist and cellular-automata researcher Stephen Wolfram proposed in 1986 — namely, to program the building blocks of a system so that they will self-assemble into whatever form you want. “Imagine you want to build a cathedral, but you don’t design a cathedral,” Mordvintsev said. “You design a brick. What shape should your brick be that, if you take a lot of them and shake them long enough, they build a cathedral for you?”

Such a brick sounds almost magical, but biology is replete with examples of basically that. A starling murmuration or ant colony acts as a coherent whole, and scientists have postulated simple rules that, if each bird or ant follows them, explain the collective behavior.  Similarly, the cells of your body play off one another to shape themselves into a single organism. NCAs are a model for that process, except that they start with the collective behavior and automatically arrive at the rules.

The possibilities this presents are potentially boundless. If biologists can figure out how Mordvintsev’s butterfly can so ingeniously regenerate a wing, maybe doctors can coax our bodies to regrow a lost limb. For engineers, who often find inspiration in biology, these NCAs are a potential new model for creating fully distributed computers that perform a task without central coordination. In some ways, NCAs may be innately better at problem-solving than neural networks.

Life’s Dreams

Mordvintsev was born in 1985 and grew up in the Russian city of Miass, on the eastern flanks of the Ural Mountains. He taught himself to code on a Soviet-era IBM PC clone by writing simulations of planetary dynamics, gas diffusion and ant colonies. “The idea that you can create a tiny universe inside your computer and then let it run, and have this simulated reality where you have full control, always fascinated me,” he said.

He landed a job at Google’s lab in Zurich in 2014, just as a new image-recognition technology based on multilayer, or “deep,” neural networks was sweeping the tech industry. For all their power, these systems were (and arguably still are) troublingly inscrutable. “I realized that, OK, I need to figure out how it works,” he said.

He came up with “deep dreaming,” a process that takes whatever patterns a neural network discerns in an image, then exaggerates them for effect. For a while, the phantasmagoria that resulted — ordinary photos turned into a psychedelic trip of dog snouts, fish scales and parrot feathers — filled the internet. Mordvintsev became an instant software celebrity.

Among the many scientists who reached out to him was Michael Levin of Tufts University, a leading developmental biologist. If neural networks are inscrutable, so are biological organisms, and Levin was curious whether something like deep dreaming might help to make sense of them, too. Levin’s email reawakened Mordvintsev’s fascination with simulating nature, especially with cellular automata.

Newswise — Fire kills nearly 3,700 Americans annually and destroys $23 billion in property, with many deaths occurring because traditional smoke detectors fail to alert occupants in time.

Now, the NYU Fire Research Group at NYU Tandon School of Engineering has developed an artificial intelligence system that could significantly improve fire safety by detecting fires and smoke in real-time using ordinary security cameras already installed in many buildings.

Published in the IEEE Internet of Things Journal, the research demonstrates a system that can analyze video footage and identify fires within 0.016 seconds per frame—faster than the blink of an eye—potentially providing crucial extra minutes for evacuation and emergency response. Unlike conventional smoke detectors that require significant smoke buildup and proximity to activate, this AI system can spot fires in their earliest stages from video alone.

“The key advantage is speed and coverage,” explained lead researcher Prabodh Panindre, Research Associate Professor in NYU Tandon’s Department of Mechanical and Aerospace Engineering (MAE). “A single camera can monitor a much larger area than traditional detectors, and we can spot fires in the initial stages before they generate enough smoke to trigger conventional systems.”

The need for improved fire detection technology is evident from concerning statistics: 11% of residential fire fatalities occur in homes where smoke detectors failed to alert occupants, either due to malfunction or the complete absence of detectors. Moreover, modern building materials and open floor plans have made fires spread faster than ever before, with structural collapse times significantly reduced compared to legacy construction.

The NYU Tandon research team developed an ensemble approach that combines multiple state-of-the-art AI algorithms. Rather than relying on a single AI model that might mistake a red car or sunset for fire, the system requires agreement between multiple algorithms before confirming a fire detection, substantially reducing false alarms, a critical consideration in emergency situations.

The researchers trained their models by building a comprehensive custom image dataset representing all five classes of fires recognized by the National Fire Protection Association, from ordinary combustible materials to electrical fires and cooking-related incidents. The system achieved notable accuracy rates, with the best-performing model combination reaching 80.6% detection accuracy.

The system incorporates temporal analysis to differentiate between actual fires and static fire-like objects that could trigger false alarms. By monitoring how the size and shape of detected fire regions change over consecutive video frames, the algorithm can distinguish between a real, growing fire and a static image of flames hanging on a wall. “Real fires are dynamic, growing and changing shape,” explained Sunil Kumar, Professor of MAE. “Our system tracks these changes over time, achieving 92.6% accuracy in eliminating false detections.”

The technology operates within a cloud-based Internet of Things architecture where multiple standard security cameras stream raw video to servers that perform AI analysis. When fire is detected, the system automatically generates video clips and sends real-time alerts via email and text message. This design means the technology can be implemented using existing CCTV infrastructure without requiring expensive hardware upgrades, an important advantage for widespread adoption.

This technology can be integrated into drones or unmanned aerial vehicles to search for wildfires in remote forested areas. Early-stage wildfire detection would buy critical hours in the race to contain and extinguish them, enabling faster dispatch of resources, and prioritized evacuation orders that dramatically reduce ecological and property loss.

To improve the safety of firefighters and assist during fire response, the same detection system can be embedded into the tools firefighters already carry: helmet cameras, thermal imagers, and vehicle-mounted cameras, as well as into autonomous firefighting robots. In urban areas, UAVs integrated with this technology can help the fire service in performing 360-degree size-up, especially when fire is on higher floors of high-rise structures.

“It can remotely assist us in confirming the location of the fire and possibility of trapped occupants,” said Capt. John Ceriello from the Fire Department of New York City.

Beyond fire detection, the researchers note their approach could be adapted for other emergency scenarios such as security threats or medical emergencies, potentially expanding how we monitor and respond to various safety risks in our society.

In addition to Panindre and Kumar, the research team includes Nanda Kalidindi (’18 MS Computer Science, NYU Tandon), Shantanu Acharya (’23 MS Computer Science, NYU), and Praneeth Thummalapalli (’25 MS Computer Science, NYU Tandon).