Model
Model explorer
AI21 Jamba 1.5 Mini
OPENAI21 Labs · AI21 Jamba family · released Aug 22, 2024
Open hybrid SSM-Transformer, 12B/52B active/total, 256K context under the Jamba Open Model License; strongest Arena-Hard score in its class.
ReasoningCodingVisionFunction callingTool useAgentic
508.6
Elo · rank #308
Parameters
52B
Active params
12B (MoE)
Context
256K tokens
Architecture
Hybrid Mamba-Transformer MoE (Jamba)
License
Jamba Open Model License
Languages
9+
API price (in/out)
$0.2 / $0.4
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
best: Llama 3.1 405B · 96.9%
best: Qwen3-235B-A22B (Non-Thinking) · 96.1%
best: ERNIE 4.5 300B-A47B · 94.3%
best: SmolLM3 3B (No Thinking) · 92.3%
best: GPT-6 Astra · 96.0%
best: Llama 3.1 405B · 96.8%
best: Claude Opus 4.5 · 99.4%
best: Gemma 4 26B A4B · 98.5%
best: Claude Fable 5 · 91.5%
best: OpenAI o3 · 92.9%
best: Gemini 3.1 Pro · 92.6%
best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%
best: Mistral Small 3.2 (24B) · 65.3%
Run it locally
VRAM @ Q4
29.5 GB
VRAM @ FP16
103.2 GB
Fits on (Q4)
RTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
GGUF Q4
Fine-tune it
Conditional / customQLoRA35.4 GB1× A100 80GB
LoRA110.8 GB1× H200 141GB
Full fine-tune834.6 GB8× H200 141GB
QLoRA SFT on ~10k samples ≈ $36.57 (1× A100 80GB)
AI21 Jamba family
Elo progression across releases
API price $0.2/$0.4 · each benchmark row carries its own source badge (see methodology)