POLL: Are any of you experiencing Tool Call Issues for the following models?

[BradKML]

I would like to note how, at a quick glance at social media, there are a lot of weird phenomena popping up:

• GLM 4.5 is surprisingly good for flexible tool calls
• Qwen3 instruct models are more compliant with new tool cal standards, while Qwen3-Coder models are made for their own tools
• DeepSeek v3.1 went the opposite direction of Qwen3, avoiding the formatting issues of the original R1 and v3
• Kimi K2 and/or Kimi Dev are being noted to be some of the worst tool-calling LLMs, but they are not overtrained to fit custom formats
• It is likely Qwen3 is inspired by both GPT-OSS models with their Harmony format, AND Anthropic's API formats
• Old names like Microsoft's Phi-4, Meta's Llama-4, and Mistral, are now behind the curve

There are no polling feature in Roo-Code Discussions, so I made it in KiloCode Kilo-Org/kilocode#2107

[BradKML]

Gonna need to add context: there has been talks around about how Qwen3-Coder being "hard to use", I would like to know if that is the case (both in BFCL and in agentic coding benchmarks).

[BradKML]

List of other places for the same question

• Are you experiencing Tool Call Issues for the following models? smallcloudai/refact#897
• Are you experiencing Tool Call Issues for the following models? bytedance/trae-agent#336
• Are you experiencing Tool Call Issues for the following models? charmbracelet/crush#1043
• Are you experiencing Tool Call Issues for the following models? TabbyML/tabby#4362
• Are any of you experiencing Tool Call Issues for the following models? continuedev/continue#7401
• Are any of you experiencing Tool Call Issues for the following models? Kilo-Org/kilocode#2107

[BradKML]

Current tally: Qwen3-Coder >> GLM-4.5 and Kimi-K2 >> DeepSeek and Qwen3-Instruct > others