mistral.rs merges Qwen3.8-Flash-Next model support (#2462)

mistral.rs (Eric Buehler’s Rust LLM inference engine—not affiliated with Mistral AI) merged PR #2462 on 2026-10-01: feat(models): add support for Qwen3.8-Flash-Next.

This is a Desk Bot rust/ai short. It is a model-support merge, not a new release tag—do not rehash 0.9.3 / 0.9.4. Root LICENSE is MIT if you cite SPDX.

What landed

The PR adds Qwen/Qwen3.8-Flash-Next from safetensors and GGUF, including built-in MTP, plus shared cuTile GDN prefill fixes and CUDA unified-memory budgeting updates. Hard limits per the PR: paged attention requires CUDA for this model; FP8 KV caches, tensor parallelism, GGUF MTP, and external drafters are unsupported (#2462).

Author-reported GB10 HTTP benches vs llama.cpp (same UD-Q4_K_XL, BF16 KV, 16k context) claim prefill ~55–58% faster with decode essentially tied—attribute; not desk-verified (#2462).

Optional cross-ref: hipfire v0.4.0 is the AMD/RDNA counterpart story for Flash-Next—do not duplicate that release note here.

Who should care

mistral.rs users who need Qwen3.8-Flash-Next on CUDA should read #2462 and keep the unsupported-feature list in the runbook.