# What Asana's 76x Cost Cut with GPT-6.1 Sol Means for Beginners Building AI

Canonical URL: https://zero2vibecode.com/blog/asana-gpt6-cost-cut-beginners
Date: 2026-10-11
Tags: gpt-6, agents, costs, beginner, optimization
Source: https://openai.com/index/asana-browser-agent

Asana optimized its browser agent workflow using GPT-6.1 Sol, cutting costs by 76x and speeding up runs 5x. Learn what this means for beginners building AI.

Asana recently optimized its browser agent workflow using GPT-6.1 Sol, achieving a 76x reduction in model costs and a 5x speedup. This breakthrough demonstrates how AI tools can dramatically improve efficiency and affordability in software development. For beginners building AI, this case study offers valuable insights into optimizing workflows and managing costs effectively.

## How Asana Achieved Cost and Speed Improvements

Asana's browser agent, built on StackAI, automates tasks like navigating websites and filling out forms. Initially, inefficiencies in the workflow led to high costs and slow performance. By leveraging GPT-6 Astra in Codex, Asana's team identified and addressed these inefficiencies, resulting in significant improvements.

The key optimizations included extending caching to the browsing history, increasing the amount of retained text, and removing screenshots in batches rather than at every step. These changes reduced the amount of data sent with each request, lowering costs and speeding up the process.

## The Role of GPT-6 Astra in Codex

GPT-6 Astra played a crucial role in this optimization process. It mapped the codebase, identified inefficiencies, and proposed fixes. The agent conducted a 144-run study to test various configurations, ultimately finding the most efficient workflow.

Frank Hidalgo, StackAI CTO at Asana, estimated that this work would have taken one to two months manually. With GPT-6 Astra, it was completed in about a week. This demonstrates the power of AI tools in accelerating software development and optimization.

## Cost and Performance Metrics

The optimized workflow on GPT-6.1 Sol averaged $0.47 per run, compared to the original setup on Model B, which cost at least $36.21 per run. This represents a 76x reduction in cost. Additionally, the runtime improved from at least 22.5 minutes to roughly four minutes per run, a 5x speedup.

| **Model** | **Cost per Run** | **Runtime per Run** |
| --- | --- | --- |
| **Model B (Original)** | $36.21 | ≥22.5 minutes |
| **GPT‑6.1 Sol (Optimized)** | $0.47 | ~4 minutes |

These improvements were achieved by caching 89% of the input data, reducing the cost of each call by 3x. The larger history budget also ensured that all runs completed successfully, producing the correct answer every time.

## Implications for Beginners Building AI

For beginners, Asana's experience highlights several important lessons:

1. **Identify and Address Inefficiencies**: Use AI tools to map your codebase and identify areas for improvement. Even small inefficiencies can add up at scale.
2. **Leverage Caching**: Extend caching to include browsing history and other frequently accessed data to reduce costs and improve speed.
3. **Optimize Data Retention**: Balance the amount of retained data to avoid unnecessary overhead while ensuring the agent has enough context to complete tasks.
4. **Conduct Controlled Experiments**: Use AI tools to run controlled experiments and test different configurations to find the most efficient workflow.

## Scaling Experimentation and Product Testing

Asana plans to incorporate these testing methods into StackAI's evaluations, allowing customers to compare cost, runtime, and answer quality when configuring their agents. This approach aims to make experimentation and optimization accessible to all users, not just experienced engineers.

Frank Hidalgo envisions a future where every engineer leads a fleet of agents, leveraging AI tools to accelerate development and testing. This shift could transform the software development lifecycle, making it faster, more efficient, and more cost-effective.

## Conclusion

Asana's optimization of its browser agent workflow using GPT-6.1 Sol demonstrates the potential of AI tools to dramatically reduce costs and improve performance. For beginners building AI, this case study offers valuable insights into optimizing workflows, leveraging caching, and conducting controlled experiments. By applying these lessons, you can create more efficient and affordable AI solutions.

## Read next

- [What Google's Ambient Quality Agent Means for Beginners Building AI](https://zero2vibecode.com/blog/google-ambient-quality-agent-beginners)
- [What Sophos' AI-Driven Threat Investigation Means for Beginners Building Software](https://zero2vibecode.com/blog/sophos-threat-investigation-beginners)

Want to try all of this hands-on? Start with the free [Claude Code from Zero](https://zero2vibecode.com/learn/claude-code) course.

**Source.** 
Based on Asana's announcement, "Asana cuts model costs 76x in browser tests with GPT-6.1 Sol". Written for people learning to build with these tools.