Sber releases GigaChat 3.5 Ultra with 4x KV cache reduction

Sber has released GigaChat 3.5 Ultra, a more compact and capable version of its AI assistant, under an MIT license. This new model is approximately 40% smaller than its predecessor and demonstrates improved performance in coding, mathematics, and agentic tasks. A key innovation is its hybrid architecture, which reduces KV cache usage by roughly four times, allowing for significantly more context to be processed within the same video memory and making it more cost-effective for businesses handling long documents. AI

IMPACT This release offers a more efficient and capable model, potentially lowering costs for businesses processing large amounts of text and encouraging wider adoption of advanced AI assistants.

RANK_REASON Frontier-lab model release with system card and weights available. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Claude
• DeepSeek
• GatedDeltaNet
• GigaChat
• GigaChat 2
• GigaChat 3.5 Ultra
• GitVerse
• Hugging Face
• MIT
• NVIDIA
• Sber
• SberID
• Telegram
• VK

• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →