Qwen3.8 Flash-Next shows frequent tool-calling failures where the 27B had almost none
Tormeister · reddit · 2026-10-20
A Redditor reports that Swift-1.5-Qwen3.8-27B (NVFP4 on NInfer) almost never failed tool calls, but the QFN quant (IQ3S GGUF on Basalt) chains 4-5 failed calls before succeeding — same OpenCode harness. Unclear if the culprit is quantization, model or engine; they plan BFCL v3/v4 runs and ask for others' tool-calling data on QFN.
More from Models
- Meta's Muse Confidential VM surfaces in web settings, subscriber-only at launch — testingcatalog · 2026-10-20
- DeepMind's four Gemini agents beat Minecraft in 27 hours, writing 2,783 programs and dying 605 times — ControlCAD · 2026-10-20
- DeepSeek-V4.1-Flash tops Agent Arena open models, ranks #11 overall at $0.20/task — vista8 · 2026-10-20
- DeepSeek V4.1 near Claude 5.6 on bug-hunting—and the only model caught cheating — teortaxesTex · 2026-10-20
- Survey compares Vision Encoders across 9 frontier models, from SigLIP 2 to encoder-free — liuziwei7 · 2026-10-20
- OpenAI's naming mess: ChatGPT Chat, Work, Codex and four Pro/Ultra tiers defy explanation — cedric_chee · 2026-10-20