Model library / ai-sage
GigaChat 3.5 Reasoning is a 438 billion-parameter mixture-of-experts language model trained with online RL for improved reasoning tasks.
8 of 256 experts active per token
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
The model is a 432 billion-parameter GigaChat 3.5 variant that combines Multi-head Latent Attention with GatedDeltaNet linear-attention layers. It is presented by ai-sage as a reasoning-focused model, trained with online reinforcement learning and domain-specific reward signals to enhance mathematics, code, instruction following and structured output.
In practice the model contains 256 experts with eight active per token, yielding 28 billion active parameters. It supports a context window of 262 144 tokens, is released under an MIT licence and uses fp8 weight format. Deployment requires substantial GPU resources, for example eight H100 GPUs, and can be run via SGLang or vLLM with optional speculative decoding.
Specification
| Repository | ai-sage/GigaChat3.5-432B-A28B-Reasoning |
| Publisher | ai-sage |
| Published | 3 September 2026 |
| Architecture | GigaChat35ForCausalLM |
| Model type | gigachat3_5 |
| Parameters | 438,085,063,424 |
| Experts | 256 total, 8 active per token |
| Layers | 40 |
| Hidden size | 7,168 |
| Attention heads | 64, 64 key/value heads |
| Context length | 262,144 tokens |
| Vocabulary | 128,256 tokens |
| Quantisation | fp8 |
| Licence | mit |
| Base model | ai-sage/GigaChat3.5-432B-A28B-base |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 816.0 GiB | Full precision as released |
| fp8 | 408.0 GiB | 8-bit, near-lossless on most models |
| int4 | 204.0 GiB | 4-bit, smallest footprint |
The cache costs 1.1 MiB per token at 16-bit, so the full 262,144-token context of one request needs about 280.0 GiB. Concurrency multiplies that number, not the weights.
That figure is an upper bound: this model uses compressed key-value attention, so the real cache is a fraction of the figure above.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | more than 8 | — | — |
| RTX 5090 32GB | more than 8 | — | — |
| L40S 48GB | more than 8 | — | — |
| A100 80GB | 8 | 140.7 GiB | about 131,758 tokens |
| H100 80GB | 8 | 140.7 GiB | about 131,758 tokens |
| RTX PRO 6000 96GB | 8 | 255.9 GiB | about 239,611 tokens |
| H200 141GB | 4 | 82.3 GiB | about 77,082 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve ai-sage/GigaChat3.5-432B-A28B-Reasoning \ --tensor-parallel-size 8 \ --max-model-len 131072 \ --gpu-memory-utilization 0.90
Questions
The weights take about 415 GiB at the released precision. No single card in the table above holds that, so it needs several GPUs or a lower precision.
The repository declares mit. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 262,144 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email [email protected] with the model name and roughly what volume you expect.
Also in the library
Nex-N2.5-Pro is a 396.8 billion-parameter MoE language model designed for long-horizon, agentic tasks with multimodal and compute…
K2-Horizon-375B-A23B is an open-weight sparse MiE language model from IFM with 379 B parameters, 23 B active per token and a 512 …
Hy4 preview is a 780 billion-parameter mixture-of-experts language model released by Tencent for productivity-focused tasks.
GLM-5.3 is a 753.3 billion-parameter open-weight mixture-of-experts language model for coding and security tasks.
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.