whatmodel.sh / GigaChat 3.5 Ultra Instruct (432B-A28B)

GigaChat 3.5 Ultra Instruct (432B-A28B)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

GigaChat 3.5 Ultra Instruct (432B-A28B) is an open-weights language model from Sber, released July 2026. Its published size is 433.77B parameters (28B active per token) and a 262k-token context window. Licence: mit.

Catalogue identity notes

ai-sage/GigaChat3.5-432B-A28B @d10dcae (created 2026-07-05, licence mit). Sber's GigaChat 3.5 Ultra instruct model (July 2026), 432B MoE with 28B active. Lab rows from the GigaChat 3.5 Reasoning card table (with Sber's run of the catalogued DeepSeek V4 Flash Preview).

Model information sources

Release date ↗ · Weights licence ↗

HF safetensors metadata 433,773,248,160 parameters; 28B active per the maker card.

55.0±1.3#267 of 719 LLMsfrom 10 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of GigaChat 3.5 Ultra Instruct (432B-A28B), out of 828 with a score, then more from Sber.

#modelscorewhereRAMreleased
319Qwen3-VL 8B Instruct Alibaba55±1api $0.31/M +local~6.8GB2025-10
320Spark-X2.5 1.7B iFLYTEK (SparkLLM)55±1local~2.7GB2026-08
321GigaChat 3.5 Ultra Instruct (432B-A28B) Sber55±1local~457GB2026-07
322Llama 3.1 405B Instruct Meta55±1api $2.40/M +local~160GB2024-07
323Qwen2.5 32B Instruct Alibaba55±1local~21GB2024-09
more from Sber
138GigaChat 3.5 Ultra Reasoning (432B-A28B) Sber67±1local~461GB2026-09

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #164 of 300 · 20% of its score · 3 tests

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 0528 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Jamba 1.5 Large here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
SWE-bench Verified lab42.6% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 31/131hard · avg 24%▇▇▇▇▇ 12%sharp · at its level · small run
Terminal-Bench 2.0 lab13.5% ✎▃▂▅▅████▆▂▃▃ beats 0/36very hard · avg 9.5%▇▇ 6.2%small run
τ³-bench lab50.0% ✎▁▁▂▁·▁▁·▁▃▂█ beats 10/25hard · avg 34%▇ 2.6%blunt · at its level · small run
coding▁▁▁▃▆██▇▆▄▃▁ #236 of 458 * · 22% of its score · 2 tests

Beats Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B), GLM-4.7-Flash here, though they rank higher overall. Trails QwQ-32B-Preview, InternLM2.5 20B Chat, Gemma 2 27B IT here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
SWE-bench Verified lab42.6% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 31/131hard · avg 24%▇▇▇▇▇ 12%sharp · at its level · small run
LiveCodeBench v6 (window unspecified) lab56.2% ✎▂▃▄▄▅██▅▄▇▃▆ beats 19/65medium · avg 46%▇▇▇▇ 11%at its level · small run
math▁▂▃▅▆▅▄█▄▁▁▁ #207 of 398 · 48% of its score · 4 tests

Beats DeepSeek V3 (Dec '24), GPT-4o (2024-08-06), Command A (03-2025) here, though they rank higher overall. Trails Olmo 3 7B Instruct, DeepSeek R1 Distill Llama 8B, Phi-4-mini-reasoning 3.8B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AIME 2026 lab67.0% ✎▁▁▁▃▁▂▆█▆▂·▁ beats 7/49medium · avg 53%▇▇▇▇▇ 14%sharp · at its level · small run
AIME 2025 lab68.0% ✎▁·▁▁▁▂▅▆█▇▅▁ beats 61/133medium · avg 48%▇▇▇▇▇ 13%sharp · at its level · small run
HMMT 2025 (edition unspecified) lab36.7% ✎▃█▇▅▇▅▁·▁··▁ beats 2/32hard · avg 30%▇▇▇▇▇ 12%sharp · at its level · small run
IMO-AnswerBench lab32.0% ✎▁▃·▁▁▁▄▅█▄▂▃ beats 2/42hard · avg 31%▇▇▇▇ 9.0%at its level · small run
other tests2 not in a task · 21% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond lab61.1% ✎▁▁▁▁▂▄▅▅███▄ beats 113/310hard · avg 45%▇▇▇▇▇▇ 16%sharp · at its level · small run
IFBench (aggregation unspecified) lab43.7% ✎▃▇▂▄▅▅▂▄▄▆█▇ beats 30/80medium · avg 40%▇▇ 4.5%at its level · small run

$ where to run it

Other ways to run it

  1. Run it as published · needs ~434 GB, NVIDIA only · from huggingface.co/ai-sage
    vllm serve ai-sage/GigaChat3.5-432B-A28B --tensor-parallel-size 8

Weights and documentation: huggingface.co/ai-sage ↗

$ how the score was worked out · ai-sage/GigaChat3.5-432B-A28B… 99%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
ai-sage/GigaChat3.5-432B-A28B-Reasoning model card · lab-reported10▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%MIT
prior · pull toward the middle▇ 1%
weighted average= 55.0

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json