Modelpublic

badtheorylabs/Tinfield-1

sourceHugging Faceotherupdated 3d agoView on Hugging Face

41likes654downloads

Model Card

Tinfield 1

Tinfield 1 is an agentic model for terminal work and long-horizon software engineering.

It scores 33.0 on Terminal-Bench 4.0, ahead of Claude Opus 4.8 at 23.6 and Claude Sonnet 5 at 12.4, and 62 on DeepSWE v1.1, ahead of Qwen3.8 Max, DeepSeek V4 Flash and Opus 4.8 again. On the Terminal-Bench board it is the second open-weight model, on a leaderboard otherwise made up entirely of frontier closed models.

It does that on 6.6B active parameters per token out of 177B total, and the quantized builds run on a single 64 GB machine.

Built on Qwen3.8-Flash-Next. 256k context, BF16 weights.

Evaluation

Benchmark	Tinfield 1	Qwen3.8-Flash-Next
Terminal-Bench 4.0	33.0	29.0
DeepSWE v1.1	62.0	58.7

Evaluated with mini-swe-agent at k=5 on the full task sets: 66 tasks for Terminal-Bench 4.0, 113 for DeepSWE v1.1.

Terminal-Bench 4.0

Model	Agent	Score
GLM-5.3 (max)	Claude Code	41.8
GPT-5.6 Sol (max)	Codex	37.3
Tinfield 1	mini-swe-agent	33.0
Qwen3.8-Flash-Next		29.0
Claude Opus 4.8 (max)	Claude Code	23.6
GPT-5.6 Terra (max)	Codex	21.5
Grok 4.6 (high)	Grok Build	20.3
Gemini 3.8 Flash (high)	mini-swe-agent	19.1
Claude Sonnet 5 (max)	Claude Code	12.4

DeepSWE v1.1

Model	Score
GLM-5.3 (max)	69
GLM-5.3 Flash (max)	63
DeepSeek V4 Pro (max)	63
Tinfield 1	62
Claude Opus 4.8 (max)	59
Qwen3.8 Max (xhigh)	57
Muse Spark 1.2 (xhigh)	55
Claude Sonnet 5 (max)	54
DeepSeek V4 Flash (max)	53

Comparison figures from the Terminal-Bench and DeepSWE leaderboards. Base model Terminal-Bench figure from Artificial Analysis; base model DeepSWE figure from the Qwen3.8-Flash-Next model card.

Quantized builds

Build	Size
Tinfield 1 Compact	72 GB	IQ2XXS gate/up, IQ4NL down
Tinfield 1 Mini	61 GB	IQ2XXS gate/up, range-searched Q20 down

Both run on a 64 GB machine. Benchmark scores above are for the BF16 weights; the quantized builds have not been evaluated on these benchmarks.

Community builds

Build	Size	By
Tinfield 1 Q4_K_XL	111 GB, 5.03 bpw	needmorevramog

Not built or measured by us. Quantizing your own is covered in QUANTIZING.md.

License

Qwen Community License 1.0, following the base model.