🎯 GigaChat 3.5 Reasoning: 5 Key Features

1️⃣ Advanced Reasoning: Explores multiple step-by-step paths, using automated verification to reinforce correct answers and self-correct

2️⃣ Autonomous Tool Usage: Independently decides when to call external APIs or revise earlier steps

3️⃣ Linear Attention: Proprietary architecture retains key context points without re-matching from scratch

4️⃣ Token Economy: Uses 37% fewer tokens than DeepSeek V4 Flash Preview on math problems

5️⃣ Proven Performance: Open-source LLM (built on GigaChat 3.5 Ultra) with massive benchmark gains:
• IFBench: 44 → 77
• Natural Plan: 64 → 80
• LiveCodeBench v6: 56 → 85

🔗 MIT License. Weights on Hugging Face:  fp8 | bf16