Cohere makes Aya 23 35B, an open-weights multilingual chat model that takes and generates text.
Input
Output
Context
32K
Max Output
-
Parameters
35B
Input Modalities
Output Modalities
Some values come from web research and may be incorrect. Report incorrect data
Download 70.0 GB · Runs on vLLM v0.32.0 · 13 GGUF builds
| Precision | Weights | One request at 32K tokens | Smallest GPU that fits |
|---|---|---|---|
| FP16 / BF16 | 65.2 GB | 73.6 GB | 1x NVIDIA H100 NVL (94 GB) |
| INT8 | 32.6 GB | 39.4 GB | 1x Moore Threads MTT S4000 (48 GB) |
| INT4 | 16.3 GB | 22.3 GB | 1x Tenstorrent Blackhole p150 (32 GB) |
Computed by Inferbase from the 34,980,831,232 parameters in the creator's Hugging Face repository; one request adds its KV cache, activations and framework overhead, with the layer layout estimated from the parameter count.
Specification sources
As of Nov 11, 2026.
Answered from the facts on this page, with their sources.
Aya 23 35B has a context window of 32,000 tokens (source: Web research). A provider may serve less; Inferbase routes on what each provider serves.
Yes. Aya 23 35B is an open-weight model under the Cc by nc 4.0 license. You can download and self-host it within the license terms.
Aya 23 35B has 34,980,831,232 parameters (35B), as stated in the creator's Hugging Face repository.
Strengths: Supports 23 languages. Limitations: License does not allow commercial use (cc-by-nc-4.0). Each is computed from the model's stated facts and benchmark results.
Aya 23 35B accepts Text and produces Text. The Capabilities card lists the API features it supports, such as function calling and structured output.
No benchmark source has published a result for Aya 23 35B yet, so it is shown as unmeasured, not estimated.
Not on Inferbase Managed today. Add your own provider key or endpoint that serves Aya 23 35B, and Inferbase runs it on your account with the same routing and logs.
The weights take 65.2 GB at FP16 / BF16, 32.6 GB at INT8, 16.3 GB at INT4, computed from the 34,980,831,232 parameters its creator publishes. Each request adds its KV cache and the framework's overhead. The hardware section shows one request and the smallest GPU that fits; the GPU planner sizes it for your context length and traffic.
The model card of Aya 23 35B lists 23 languages: English, French, German, Spanish, Italian, Portuguese, Japanese, Korean, Chinese, Arabic, Greek, Persian, Polish, Indonesian, Czech, Hebrew, Hindi, Dutch, Romanian, Russian, Turkish, Ukrainian and Vietnamese. That is what Cohere states; the catalog does not test language quality.
Start on the free plan with 5,000 requests a month, no card required.