HaiAI123

Curated Global AI Tools Directory

Industry update
Flat

POCKET-27B-Pi coding model released with 16.5GB quantized build

11/21/2026 — 11/21, 08:18·1 sources (all)·1 reports (all)

Key takeaways

  • A coding-focused model called POCKET-27B-Pi is now public on Hugging Face, per DEV Community. Its Q4_K_M quantized file is 16.5GB and runs o…
  • it solved 28 of 48 shared Terminal-Bench 2.1 tasks (58.3%), ahead of the original Qwen3.8-27B's 19 and the 53.8GB full-precision pi model's…

Background & analysis

According to a DEV Community post published at 07:32 on 2026-11-21, the coding-focused model POCKET-27B-Pi has been released on Hugging Face under the Apache-2.0 license, with the model file available at huggingface.co/FINAL-Bench/POCKET-27B-Pi-GGUF.

The Q4_K_M quantized file of POCKET-27B-Pi weighs 16.5GB and can run on a single 24GB graphics card, with the file roughly one third the size of the full-precision release. On the 48 shared tasks of Terminal-Bench 2.1, the quantized build solved 28 tasks, or 58.3%. For comparison, the original Qwen3.8-27B solved 19 tasks (39.6%), and the 53.8GB full-precision pi model solved 21 tasks (43.8%). The post states this is an improvement of 47% over the original weights and 33% over the full-precision pi model. Under the same 50-task configuration, the median number of agent steps per task for the quantized build was 15, lower than the 36.5 steps recorded for the original weights, and no run reached the 100-step cap.

As of now, the public record of this release consists of the model going live on Hugging Face together with the benchmark figures described above; no further developments are reported in the available coverage.

AI-generated from 1 reports · updated 1 day ago

Related AI tools

updream
24-hour heatHeat index 46 · peak 90 · 23h ago
24 hours agonow

Reports on this story headlines open the original

Yesterday
  1. A DEV Community post highlights POCKET-27B-Pi, a coding-focused model released on Hugging Face. Its Q4_K_M file is 16.5GB and runs on a single 24GB GPU; it solved 28 of 48 shared Terminal-Bench 2.1 tasks (58.3%), versus 19 for the original Qwen3.8-27B and 21 for the 53.8GB full-precision pi model.

    DEV Community · AIAI score 59

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →