262K $0.2 / $1.15 395.6 н/д cloud + OW text · vision gpt-oss-120b (high) reasoning OpenAI н/д $0.15 / $0.6 249.8 70 GB · Q4_K weights ≈3× RTX 4090 · ≈2× RTX 6000 Ada (48 GB) cloud + OW text · code 256K $0.1 / $0.8 247.5 н/д cloud text · vision · audio · video 256K $0.2 / $0.8 224.1 н/д cloud text · code GPT-5.6 Luna (max) reasoning OpenAI 1M $1 / $6 206.9 н/д cloud text · vision · code 1M $2.5 / $7.5 199.8 80 GB · Q4 MoE weights ≈4× RTX 4090 · ≈2× RTX 6000 Ada open-weight text · vision · code Gemini 3.5 Flash Lite reasoning Google 1M $0.4 / $1.6 192.5 н/д cloud text · vision · code Nemotron 3 Ultra reasoning NVIDIA 262K $0.675 / $2.67 187.9 160 GB · Q4 MoE 550B-A55B ≈7× RTX 4090 · ≈4× RTX 6000 Ada · ≈2× H100 80GB open-weight text · code 256K $0.3 / $1 181.7 24 GB · Q4_K weights (35B-A3B) ≈1× RTX 4090 · ≈1× RTX 6000 Ada open-weight text · vision · code 1.0M $0.1 / $0.4 170 н/д cloud text · vision · audio · video · code Gemini 3.5 Flash reasoning Google 1M $1.5 / $9 159.2 н/д cloud text · vision · code 128K $0.2 / $0.6 158.4 16 GB · Q4_K weights (24B) ≈1× RTX 4090 · ≈1× RTX 3090 open-weight text · vision · code 1.0M $0.4 / $1.6 150 н/д cloud text · vision · code 500K $0.8 / $4 148.6 н/д cloud text · vision · code GPT-5.4 (xhigh) reasoning OpenAI 1.1M $2.5 / $15 145.9 н/д cloud text · vision · code Nex-N2-Pro reasoning Nex 262K $0.5 / $2.5 142.4 н/д cloud text · code GLM-5.2 (max) reasoning Z.ai 1M $1.4 / $4.4 141.5 90 GB · Q4 / mixed MoE ≈4× RTX 4090 · ≈2× RTX 6000 Ada open-weight text · code 400K $0.25 / $2 140 н/д cloud text · vision · code GPT-5.6 Terra (max) reasoning OpenAI 1M $2.5 / $15 137.3 н/д cloud text · vision · code 1M $0.14 / $0.28 132.8 50 GB · Q4 MoE (284B) ≈3× RTX 4090 · ≈2× RTX 6000 Ada open-weight text · code 256K $0.6 / $2.5 122.7 н/д cloud + OW text · code Gemini 2.5 Flash reasoning Google 1.0M $0.3 / $2.5 120 н/д cloud text · vision · audio · video · code 1M $0.3 / $2.5 120 н/д cloud text · vision · video Muse Spark 1.1 (xhigh) reasoning Meta 1.0M $1.25 / $4.25 117.6 48 GB · Q4_K weights ≈2× RTX 4090 · ≈1× RTX 6000 Ada open-weight text · code Grok 4.5 (high) reasoning xAI 500K $2 / $6 116.3 н/д cloud text · vision · code Gemini 3.1 Pro Preview reasoning Google 1M $2 / $12 115.9 н/д cloud text · vision · code 256K $0.35 / $1.2 114.3 н/д cloud + OW text · code 256K $2.5 / $10 110.2 н/д cloud text · code GLM-4.7 Flash reasoning Z.ai 200K $0.07 / $0.4 103.2 н/д cloud + OW text · code 2M $2.5 / $15 102.4 н/д cloud text · vision · audio · video · code 256K $2 / $6 96.5 н/д cloud text · vision · code Grok 4.3 reasoning xAI 1M $1.25 / $2.5 95 н/д cloud text · vision · code 256K $1.5 / $5 92.3 60 GB · Q4 MoE weights ≈3× RTX 4090 · ≈2× RTX 6000 Ada open-weight text · vision · code 1.0M $2 / $8 90 н/д cloud text · vision · code 205K $0.3 / $1.2 85.1 н/д cloud + OW text · code 1M $3 / $15 85 н/д cloud text · vision · code 1M $0.3 / $1.2 83.7 40 GB · Q4_K (если open-веса) ≈2× RTX 4090 · ≈1× RTX 6000 Ada cloud text · code 256K $1 / $4 81.2 70 GB · Q4 MoE weights ≈3× RTX 4090 · ≈2× RTX 6000 Ada open-weight text · vision · audio · video · code GLM-5.1 reasoning Z.ai 200K $1.4 / $4.4 81 н/д cloud + OW text · code 1M $3 / $15 80 н/д cloud text · vision · code 400K $1.25 / $10 75 н/д cloud text · vision · code 1M $1.87 / $4.68 72.7 48 GB · Q4_K weights (оценка) ≈2× RTX 4090 · ≈1× RTX 6000 Ada open-weight text · code 1M $2 / $10 71.3 н/д cloud text · vision · code GPT-5.5 (xhigh) reasoning OpenAI 922K н/д 66.3 н/д cloud text · vision · code Claude Fable 5 (with fallback) reasoning Anthropic 1M н/д 63.9 н/д cloud text · vision · code 262K $0.6 / $3.6 63.7 н/д cloud + OW text · vision 1M $0.435 / $0.87 62.3 80 GB · Q4 MoE (full expert set) ≈4× RTX 4090 · ≈2× RTX 6000 Ada open-weight text · code 1M $0.14 / $0.28 62.3 н/д cloud + OW text · vision 256K $0.14 / $0.58 60.3 н/д cloud + OW text · code 200K $1.1 / $4.4 60 н/д cloud text · vision · code GPT-5.6 Sol (max) reasoning OpenAI 1M н/д 56.7 н/д cloud text · vision · code 262K $0.6 / $3.6 56.1 н/д cloud + OW text · vision · video 1.0M $1.25 / $10 55 н/д cloud text · vision · audio · video · code 1M $0.4 / $1.6 53.3 н/д cloud text · vision · video 256K $0.4 / $4.8 53.3 н/д cloud text · vision · audio · video 1M $2.5 / $13 50 н/д cloud text · vision 256K $0.95 / $4 47.6 н/д cloud + OW text · vision · video · code 256K $0.95 / $4 44.4 н/д cloud text · code 1M $3 / $15 42 180 GB · Q4 MoE 2.8T (оценка) ≈8× RTX 4090 · ≈4× H100 80GB cloud + OW text · vision · code 200K $2 / $8 40 н/д cloud text · vision · code 200K $15 / $75 30 н/д cloud text · vision · code 1M $5 / $25 н/д н/д cloud text · vision · code Qwen3.8 Max Preview reasoning Alibaba 984K н/д н/д 150 GB · Q4 MoE 2.4T (оценка, когда OW) ≈7× RTX 4090 · ≈4× RTX 6000 Ada cloud text · vision · code 1M $0.435 / $0.87 н/д н/д cloud text · vision · code Llama 4 Maverick Meta 1M н/д н/д 100 GB · Q4 MoE weights ≈5× RTX 4090 · ≈2× RTX 6000 Ada open-weight text · vision · code 512K н/д н/д 40 GB · Q4_K weights ≈2× RTX 4090 · ≈1× RTX 6000 Ada open-weight text · vision · code 128K $0.28 / $0.42 н/д н/д cloud + OW text · code 128K н/д н/д 380 GB · Q4 MoE (671B-A37B) ≈8× H100 80GB · ≈16× RTX 4090 open-weight text · code GLM-5V Turbo reasoning Z.ai 200K н/д н/д н/д cloud text · vision · video 1M н/д н/д н/д cloud + OW text · code 262K $0.25 / $2 н/д н/д cloud text · vision · video 128K н/д н/д 40 GB · Q4_K weights (70B) ≈2× RTX 4090 · ≈1× RTX 6000 Ada open-weight text · code 128K н/д н/д 15 GB · Q4_K weights (24B) ≈1× RTX 4090 · ≈1× RTX 3090 open-weight text · vision · code 128K н/д н/д 9 GB · Q4_K weights (14B) ≈1× RTX 3090 / 4070 Ti Super open-weight text · vision · code 256K н/д н/д 14 GB · Q4_K weights (22B) ≈1× RTX 4090 · ≈1× RTX 3090 open-weight text · code 128K н/д н/д 15 GB · Q4_K weights (24B) ≈1× RTX 4090 open-weight text · code 128K н/д н/д 20 GB · Q4_K weights (32B) ≈1× RTX 4090 · ≈1× RTX 6000 Ada open-weight text · code 128K н/д н/д 6 GB · Q4_K weights (8B) ≈1× RTX 3060 12GB · Mac M-серия open-weight text · code 128K н/д н/д 20 GB · Q4_K weights (32B) ≈1× RTX 4090 open-weight text · code 128K н/д н/д 18 GB · Q4 MoE (30B-A3B) ≈1× RTX 4090 open-weight text · code 128K н/д н/д 135 GB · Q4 MoE (235B-A22B) ≈6× RTX 4090 · ≈2× H100 80GB open-weight text · code 256K н/д н/д 270 GB · Q4 MoE (480B-A35B) ≈4× H100 80GB · ≈8× RTX 6000 Ada open-weight text · code 128K н/д н/д 42 GB · Q4_K weights (72B) ≈2× RTX 4090 · ≈1× RTX 6000 Ada open-weight text · vision 32K н/д н/д 6 GB · Q4_K weights (7B) ≈1× RTX 3060 12GB open-weight text · vision · audio · video 128K н/д н/д 380 GB · Q4 MoE (671B-A37B) ≈8× H100 80GB · ≈16× RTX 4090 open-weight text · code DeepSeek-R1-Distill Qwen 32B reasoning DeepSeek 128K н/д н/д 20 GB · Q4_K weights (32B) ≈1× RTX 4090 open-weight text · code 128K н/д н/д 17 GB · Q4_K weights (27B) ≈1× RTX 4090 open-weight text · vision 128K н/д н/д 16 GB · Q4 MoE (21B-A3.6B) ≈1× RTX 4090 · ≈1× RTX 3090 open-weight text · code 128K н/д н/д 8 GB · Q4_K weights (12B) ≈1× RTX 3060 12GB open-weight text · code