Skip to content
NewA calmer, faster ZeroHour
StoryModel releaseinfo

Recursive’s Flint coding model trends on Hugging Face

1 source · 2 articlesFirst reported Updated ()
What's new

October 21: RecursiveSI/flint at rank 29 adds a BF16 checkpoint, SGLang kit, Apache 2.0 license, about 80 GB download, and a remaining-risk safety note.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Recursive’s Flint coding-agent model is trending on Hugging Face with Apache 2.0 weights and developer-reported scores.

Recursive published Flint, a mixture-of-experts model for coding agents, and related listings are trending on Hugging Face. Flint has 39.8 billion parameters, about 2.5 billion active per token, 49 layers, and a 262,144-token context, with a stated June 2026 knowledge cutoff. Recursive reports 54.9 percent on Terminal-Bench 2.1, 89.7 percent on SWE-bench Pro v2, and 74.2 percent on SWE-bench Verified; it says setups differ and some comparison scores are developer-reported. The RecursiveSI checkpoint is BF16 with an SGLang kit, Apache 2.0 weights, and a documented one-GPU serving setup of about 80 GB. Safety fine-tuning is included, but the card says harmful output and unsafe tool calls remain possible.

Key facts
  • 39.8 billion total parameters; about 2.5 billion active per token.
  • 49 layers, 256 routed experts, six selected per token, plus a shared expert.
  • Documented context is 262,144 tokens; stated knowledge cutoff is June 2026.
  • Weights are Apache 2.0; hosted API terms are separate.
  • One B300 or H200 GPU per replica; download is about 80 GB.
  • Recommended sampling is temperature 1.0, top-p 0.95, and top-k 20.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

Timeline

dated events in the reporting · oldest first
  1. Oct 20, 2026unsloth/flint-GGUF trends at number 30 · Hugging Face trending models
  2. Oct 21, 2026RecursiveSI/flint trends at number 29 · Hugging Face trending models

What each outlet adds

beyond the shared facts
  • Hugging Face trending modelsGGUF listing, 74.2% SWE-bench Verified, shared expert, sampling settings.
  • Hugging Face trending modelsBF16 and SGLang release, Apache 2.0, GPU size, safety limits.
Vendors
Recursive
Organizations
RecursiveUnsloth
AI models
Flint

Coverage

oldest first · one row per article · UTC
  1. · 5d ago Hugging Face trending modelsFirst report

    Unsloth lists GGUF builds of Recursive's Flint coding model

    Flint is a 39.8B MoE model with about 2.5B active parameters, scored by Recursive at 54.9% on Terminal-Bench 2.1.

  2. · 4d ago Hugging Face trending models

    Recursive releases Flint, a 39.8B coding-agent model under Apache 2.0

    About 2.5 billion parameters are active per token; Recursive reports 54.9% on Terminal-Bench 2.1 and 89.7% on SWE-bench Pro v2.