Playwright not executed when using together.ai models like meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8

[roman-doubrava]

What happened?

When using tool use like playwright & filesystem with together.ai models like meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8 the tools are ignored. Example prompt is "Go to a website and use tool playwright to browse and make screenshots ans save the screenshots.", the model gives back "Your input is not sufficient. Please provide more details or specify the task you need help with." wihtout calling any tool.

The same prompt works with anthropic claude 3.7+ without issues.

Version Information

a2f330e

Steps to Reproduce

1. configure Togeteher.ai
2. Configure MCP tools filesystem + playwright
3. Select model meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
4. Prompt: "Webscrape the website XXX and tell me about their products, make screenshots of each product side and create a HTML Summary file with the screenshots and the product descriptions. Use tool playwright to scrape the website."
5. Result no tool use, or "Your input is lacking necessary details. Please provide more information or specify the task you need help with."

What browsers are you seeing the problem on?

Chrome

Relevant log output

2025-06-10T07:54:46.167Z debug: [spendTokens] No transactions incurred against balance
2025-06-10T07:54:46.171Z debug: [saveConvo] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2025-06-10T07:54:46.189Z debug: [AgentController] Request closed
2025-06-10T07:54:46.193Z debug: [AgentController] Performing cleanup
2025-06-10T07:54:46.193Z debug: [AgentController] Cleaning up abort controller
2025-06-10T07:54:46.194Z debug: [AgentController] Cleanup completed

Screenshots

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Duplicate: #7818