Model
Model explorer
Koala 13B
OPENUC Berkeley (BAIR) · Koala family · released Apr 3, 2023
LLaMA 13B fine-tuned on curated web dialogue (ShareGPT + HC3); a parallel academic effort to Vicuna, often preferred over Alpaca in blind human evals.
ReasoningCodingVisionFunction callingTool useAgentic
-4.1
Elo · rank #405
Parameters
13B
Active params
13B (dense)
Context
2K tokens
Architecture
LLaMA (dense decoder-only transformer)
License
Research-only: LLaMA non-commercial license (base weights) + OpenAI ToS (ShareGPT-derived data); training/serving code Apache-2.0
Languages
—
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
best: Llama 3.1 405B · 96.9%
best: Claude Fable 5 · 1505
best: Hunyuan-T1 · 93.1
best: Llama 3.1 405B · 96.8%
best: Claude 3 Opus · 95.4%
best: OpenAI o3 · 92.9%
best: Hunyuan-Large (A52B) · 9.4
best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%
best: PaLM 2 · 90.9%
Run it locally
VRAM @ Q4
8 GB
VRAM @ FP16
26 GB
Fits on (Q4)
RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
13B at Q4 fits comfortably in 24GB VRAM; no model-specific RTX 4090 llama.cpp measurement found, though similarly-sized Llama models typically reach 60-90 tok/s.
Quantizations
GPTQ · GGML
API price weights · each benchmark row carries its own source badge (see methodology)