Skip to content
Updated minutes ago
Cohere

Aya 23 35B

Cohere · dense · 35B parameters · 131,072 context

Quality
50.0

Parameters

35B

Context Window

128K tokens

Architecture

Dense

Best GPU

H20

Cheapest API

$1.50/M

Intelligence Brief

Aya 23 35B is a 35B parameter DENSE model from Cohere, featuring Grouped Query Attention (GQA) with 32 layers and 8,192 hidden dimensions. With a 131,072 token context window, it supports multilingual. The most cost-effective API deployment is via cohere at $1.50/M output tokens. For self-hosted inference, H20 delivers optimal throughput at $940/month.

Architecture Details

TypeDENSE
Total Parameters35B
Active Parameters35B
Layers32
Hidden Dimension8,192
Attention Heads64
KV Heads8
Head Dimension128
Vocab Size256,000

Memory Requirements

BF16 Weights

70.0 GB

FP8 Weights

35.0 GB

INT4 Weights

17.5 GB

KV-Cache per Token131072 bytes
Activation Estimate2.00 GB

GPU Compatibility Matrix

Aya 23 35B is compatible with 57% of GPU configurations across 41 GPUs at 3 precision levels.

BF16 (Full)
FP8 (Half)
INT4 (Quarter)
Blackwell(7 GPUs)
B200 NVL (pair)360GB
B300288GB
B100 SXM192GB
GB200 NVL72 (per GPU)192GB
Hopper(7 GPUs)
H100 NVL 94GB (per GPU pair)188GB
H200 SXM141GB
H2096GB
GH20096GB
Ada Lovelace(11 GPUs)
L40S48GB
L4048GB
RTX 6000 Ada48GB
L2048GB
Ampere(16 GPUs)
A100 80GB SXM80GB
A100 80GB PCIe80GB
A1664GB
RTX A600048GB
Legend:No fitVery tightTightModerateGoodExcellent

GPU Recommendations

H20optimal

FP8 · 1 GPU · tensorrt-llm

100/100

score

Throughput

956.1 tok/s

Latency (ITL)

1.0ms

Est. TTFT

0ms

Cost/Month

$940

Cost/M Tokens

$0.37

Use this config →
B200 SXMoptimal

FP8 · 1 GPU · tensorrt-llm

98/100

score

Throughput

1.1K tok/s

Latency (ITL)

1.0ms

Est. TTFT

0ms

Cost/Month

$4261

Cost/M Tokens

$1.54

Use this config →
H200 SXMoptimal

FP8 · 1 GPU · tensorrt-llm

95/100

score

Throughput

1.1K tok/s

Latency (ITL)

1.0ms

Est. TTFT

0ms

Cost/Month

$2553

Cost/M Tokens

$0.93

Use this config →

Deployment Options

API

API Deployment

cohere

$1.50/M

output tokens

Self-Hosted

Single GPU

H20

$940/mo

Min VRAM: 35 GB

Scale

Multi-GPU

RTX A6000 x2

104.7 tok/s

TP· $930/mo

API Pricing Comparison

ProviderInput $/MOutput $/MBadges
cohere$0.50$1.50
Cheapest

Cost Analysis

ProviderInput $/MOutput $/M~Monthly Cost
cohereBest Value$0.50$1.50$10

Cost per 1,000 Requests

Short (500 tok)

$0.55

via cohere

Medium (2K tok)

$2.20

via cohere

Long (8K tok)

$7.00

via cohere

Performance Estimates

Throughput by GPU

H20
956.1 tok/s
B200 SXM
1.1K tok/s
H200 SXM
1.1K tok/s

VRAM Breakdown (H20, FP8)

Weights
Act
Weights 35.0 GBKV-Cache 1.1 GBActivations 16.0 GBOverhead 1.8 GB

Precision Impact

bf16

70.0 GB

weights/GPU

fp8

35.0 GB

weights/GPU

~956.1 tok/s

int4

17.5 GB

weights/GPU

Capabilities

Features

Tool Use Vision Code Math Reasoning Multilingual Structured Output

Supported Frameworks

vllmsglangtgitensorrt-llm

Supported Precisions

BF16 (default)FP8INT4

Where to Deploy Aya 23 35B

Similar Models

Command R

35B params · dense

Quality: 68

from $0.50/M

Higher quality, CheaperCompare →
Higher quality, CheaperCompare →

Yi 1.5 34B

34.4B params · dense

Quality: 72

from $0.80/M

Larger context, Higher quality, CheaperCompare →

Code Llama 34B

34B params · dense

Quality: 55

from $0.78/M

Smaller context, CheaperCompare →

Frequently Asked Questions

How much VRAM does Aya 23 35B need for inference?

Aya 23 35B requires approximately 70.0 GB of VRAM at BF16 precision, 35.0 GB at FP8, or 17.5 GB at INT4 quantization. Additional VRAM is needed for KV-cache (131072 bytes per token) and activations (~2.00 GB).

What is the best GPU for Aya 23 35B?

The top recommended GPU for Aya 23 35B is the H20 using FP8 precision. It achieves approximately 956.1 tokens/sec at an estimated cost of $940/month ($0.37/M tokens). Score: 100/100.

How much does Aya 23 35B inference cost?

Aya 23 35B API inference starts from $0.50/M input tokens and $1.50/M output tokens. Self-hosted inference costs depend on your GPU configuration — use our ROI calculator for a detailed breakdown.