GigaChat 3.5 Reasoning: Advanced step-by-step problem solving GigaChat 3.5 Reasoning is a new open-source LLM designed to reason, not just respond

🚀 GigaChat 3.5 Reasoning — a new open-source LLM that thinks before it answers. It breaks problems into stages, builds a plan, checks intermediate results, and self-corrects. Built on GigaChat 3.5 Ultra, it explores multiple step-by-step reasoning paths for math & coding, using automated verification to reinforce correct answers. ⚡️ Proprietary linear attention makes it highly efficient on long contexts, retaining key points without re-matching from scratch. It’s also token-efficient: uses 37% fewer tokens than DeepSeek V4 Flash Preview on math problems! 📈 Benchmark gains over non-reasoning version: • IFBench: 44 → 77 • Natural Plan: 64 → 80 • LiveCodeBench v6: 56 → 85 📦 MIT license. Weights on Hugging Face: fp8 | bf16

❓ Tired of LLMs that hallucinate on complex tasks? What if your AI could: • Explore multiple step-by-step reasoning paths? • Use automated verification to reinforce correct answers? • Check its own work and self-correct? • Autonomously decide when to call external tools? GigaChat 3.5 Reasoning does all of this. This open-source LLM (built on GigaChat 3.5 Ultra) actually thinks before answering. It uses proprietary linear attention for efficient long-context handling, consuming 37% fewer tokens than DeepSeek V4 Flash Preview on math problems. Real-world performance: ✓ IFBench: 44 → 77 ✓ Natural Plan: 64 → 80 ✓ LiveCodeBench v6: 56 → 85 🔗 MIT License. Weights on Hugging Face: fp8 | bf16

GigaChat 3.5 Reasoning: Advanced step-by-step problem solving GigaChat 3.5 Reasoning is a new open-source LLM designed to reason, not just respond. Built on GigaChat 3.5 Ultra, it processes math and coding tasks by exploring multiple step-by-step reasoning paths, using automated verification to reinforce correct trajectories. This enables autonomous multi-step planning, external tool invocation, and self-correction. The model features a proprietary linear attention architecture, optimizing long-context efficiency by retaining key processed points rather than re-matching from scratch. It is highly token-efficient, consuming on average 37% fewer tokens than DeepSeek V4 Flash Preview on math problems. Benchmark gains over the non-reasoning version: • IFBench: 44 → 77 • Natural Plan: 64 → 80 • LiveCodeBench v6: 56 → 85 Open-sourced under the MIT license. Weights available on Hugging Face: fp8 | bf16

🤖 New Powerful AI Model: GigaChat 3.5 Reasoning This open-source LLM actually thinks before it answers! Perfect for complex coding, math, and reasoning prompts. ✅ Built on GigaChat 3.5 Ultra: explores multiple step-by-step reasoning paths ✅ Automated verification reinforces correct answers, enabling self-correction ✅ Autonomously decides when to call external tools or revise earlier steps ✅ Highly efficient: Linear attention uses 37% fewer tokens than DeepSeek V4 Flash Preview 📈 Massive benchmark gains over non-reasoning versions: • IFBench: 44 → 77 • Natural Plan: 64 → 80 • LiveCodeBench v6: 56 → 85 🔗 Open-sourced under MIT license. Weights on Hugging Face: fp8 | bf16