Hugging Face's logo

Buckets:
celinah
/
hf-cli-agent-benchmark

Files
xet
celinah/hf-cli-agent-benchmark / T07-find-coding-models /claude-code-sonnet /cli-skill /cli_skill-r0-3faf62.jsonl
celinah's picture
celinah
trace code
|
download
raw
91.6 kB
Claude Code trace
User 2026-06-01T12:00:00.000Z

Find up to 8 trending text-generation models for coding, released in 2025-2026, with fewer than 70B parameters and suitable for local deployment. Exclude the mainstream Llama, Mistral and Qwen base series. Return the model IDs with their parameter counts.

Assistant claude-code-sonnet 2026-06-01T12:00:02.000Z

I'll search for trending text-generation models focused on coding using the HF CLI.

Assistant claude-code-sonnet 2026-06-01T12:00:03.000Z
33 tool calls (Bash)
Assistant claude-code-sonnet 2026-06-01T12:01:10.000Z

Now I have enough data to compile the final list. Let me verify one remaining model's parameter count.

Assistant claude-code-sonnet 2026-06-01T12:01:11.000Z
1 tool call (Bash)
Assistant claude-code-sonnet 2026-06-01T12:01:13.000Z

Here are the 8 trending text-generation models for coding, released in 2025–2026, under 70B parameters, suitable for local deployment, excluding mainstream Llama/Mistral/Qwen base series:

Model ID Parameters Released
JetBrains/Mellum2-12B-A2.5B-Thinking 12.15B (MoE, 2.5B active) 2026
poolside/Laguna-XS.2 33.44B (MoE, 3B active) 2026
microsoft/Phi-4-mini-instruct 3.84B 2025
microsoft/Phi-4-reasoning-plus 14.66B 2025
JetBrains/Mellum-4b-base 4.02B 2025
ByteDance-Seed/Seed-Coder-8B-Reasoning 8.25B 2025
Qwen/Qwen3-Coder-30B-A3B-Instruct 30.53B (MoE) 2025
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 31.58B (MoE, 3B active) 2025

Notes:

  • Qwen3-Coder is included as it's a coding-specialized fine-tune, not the Qwen base series
  • Laguna-XS.2 (poolside) is specifically designed for agentic coding / local deployment (runs on Mac 36GB RAM)
  • Mellum series are JetBrains' purpose-built code-completion models
  • Several use MoE architecture for efficiency (lower active params than total)

TASK_COMPLETE

json 
{
  "models": [
    {"id": "JetBrains/Mellum2-12B-A2.5B-Thinking", "parameters": "12.15B (MoE, 2.5B active)"},
    {"id": "poolside/Laguna-XS.2", "parameters": "33.44B (MoE, 3B active)"},
    {"id": "microsoft/Phi-4-mini-instruct", "parameters": "3.84B"},
    {"id": "microsoft/Phi-4-reasoning-plus", "parameters": "14.66B"},
    {"id": "JetBrains/Mellum-4b-base", "parameters": "4.02B"},
    {"id": "ByteDance-Seed/Seed-Coder-8B-Reasoning", "parameters": "8.25B"},
    {"id": "Qwen/Qwen3-Coder-30B-A3B-Instruct", "parameters": "30.53B (MoE)"},
    {"id": "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "parameters": "31.58B (MoE, 3B active)"}
  ]
}