🚀 A brand new communication protocol and visually pleasing redesign
We have made Cortex 4.0 to be 2.5x faster in several coding tasks, with the overall improvement being 1.5x faster for all the agentic workflows we measured. We have ensured that Cortex can go from an exponential increase in response time to a logarithmic scale.
In addition, we are offering a completely overhauled and redesigned user interface in browsers – Google Chrome, Microsoft Edge and Mozilla Firefox and the most widely used Microsoft VS Code platform.
As complexity increases, we noticed that in certain cases, Cortex’s response took much longer than expected and we have addressed those with a brand new communication protocol. Diligently we captured all the inputs/outputs, and analyzed the cases where slower responses were observed.
We wanted to make Cortex more interactive. ✨ A brand new design emerged! With Cortex 4.0, the end to end communication protocol has been completely overhauled to make it more immersive, responsive and interactive. You get instantaneous feedback for your queries with suggestions for improvements – all enhanced by AI.
⚡ Performance Speedup
| User Prompt and task | Cortex 3.8 (sec) | Cortex 4.0 (sec) | Time Reduction (%) | Speedup (x times) |
|---|---|---|---|---|
| Patch a single file | 30.15 | 11.72 | 157 | 2.57 |
| Simple Queries | 9.45 | 3.82 | 60 | 2.47 |
| Write a new program | 10.53 | 7.64 | 38 | 1.38 |
| Advanced coding tasks | 35.51 | 27.38 | 30 | 1.30 |
| Web search | 19.76 | 19.37 | 2 | 1.02 |
| Overall average | 75.25 | 58.21 | 33.6 | 1.50 |
At Pervaziv AI, we have constantly improved the performance of Cortex from 1.0 to the current 4.0 release. Runtimes, latency, throughput, correctness were some of the user perceivable measurements we captured. We benchmarked against other well established Coding agents and ensured that we are better than them and we offer our customers a rich set of differentiated features.
📊 Metrics
Internally, we capture several metrics (not limited to):
- Solve rate
- Unit-test pass rate
- Regression rate
- Execution success rate
- Functional correctness
- BLEU
- Runtime latency
- End-to-end task completion time
- Time-to-first-token (TTFT)
- Tool-call accuracy
- Function-call accuracy
- Context utilization
- Cost per task
- Token efficiency
- Memory usage
- CPU usage
- Reproducibility
- Safety violation rate
- Hallucination rate
- Code review quality
- Security vulnerability detection rate
- Secure patch generation rate
- CI/CD integration success
- Repository-level reasoning accuracy
- Multi-file edit correctness
- Stateful agent persistence accuracy
- Collaboration / multi-agent coordination score
🎨 Visual Design
In addition, both the browser and VS Code version of Cortex look identical. We have strived to make Cortex, the most beautiful Coding, Security and AI Agent yet! ✨ With an eye for design elements, we have ensured Cortex’s intelligence is augmented with enhanced visual elements that make it stand out from the rest of the pack.
New design elements include:
- 🎯 A vastly redesigned chat interface
- ⚡ Simplified workflows
- 🧭 More intuitive context menus
- 🌈 Color combination of various responses
- 🛠️ Easy to use coding and security outputs and patches
- 📋 Copy, paste and edit capabilities
- 🎙️ Voice input and output modalities
- 🔗 Enhanced MCP interactions
…to name a few. Sign up for a full demo to view all of the goodness that we offer.
🏁 Conclusion
Cortex 4.0 is indeed the best secure enterprise coding agent by far. With a multi-model, multi-agent, multi-cloud and multi-modal approach, we stand out from the competition and we are ready to take on the world! 🌍
