Model
Model explorer

Aya 23 35B

OPEN
Cohere Labs · Aya 23 family · released May 23, 2024

Cohere's 35B dense multilingual instruct model spanning 23 languages, research-only CC-BY-NC release.

ReasoningCodingVisionFunction callingTool useAgentic
228.8
Elo · rank #365
Parameters
35B
Active params
35B (dense)
Context
8K tokens
Architecture
Dense transformer (Command-based), 35B
License
CC-BY-NC
Languages
23+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ARC-ChallengeReasoning61.7%#84
best: Llama 3.1 405B · 96.9%
BIG-Bench HardReasoning53.6%#89
best: ERNIE 4.5 300B-A47B · 94.3%
GPQA DiamondReasoning28.8%#273
best: GPT-6 Astra · 96.0%
IFEvalReasoning59.3%#123
best: Gemma 4 26B A4B · 98.5%
LogicKorHuman preference6.76#21
best: GPT-4o · 9.33
MGSMMath53.7%#44
best: OpenAI o4-mini · 93.7%
MMLU-ProKnowledge33.6%#162
best: Claude Fable 5 · 91.5%
MMLUKnowledge58.2%#210
best: OpenAI o3 · 92.9%
XWinogradReasoning84.4%#3
best: Sarashina2-70B · 91.8%
Run it locally
VRAM @ Q4
22 GB
VRAM @ FP16
70 GB
Fits on (Q4)
RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
GGUF Q4
Fine-tune it
Research only
QLoRA24.1 GB1× RTX 5090 32GB
LoRA74.8 GB1× A100 80GB
Full fine-tune562.0 GB4× H200 141GB
QLoRA SFT on ~10k samples ≈ $17.07 (1× RTX 5090 32GB)
Plan a fine-tune →
Aya 23 family
Elo progression across releases
Aya 23 35BMay 2024228.8
Resources
HFCohereLabs/aya-23-35BDOCSdocs.cohere.com/docs/aya
API price weights · each benchmark row carries its own source badge (see methodology)