便宜又好用的 LLM 怎么选？
跑量场景不只看单价，要看「10 美元能跑多少 token」。我们把输出价 ≤ $3.5/M 的模型拉出来，按价格、智能、速度三维排个序。
价格数据 2026-07-22 · 智能分 2026-07-21
经济型模型价格总表（按输出价升序）
| 模型 | 输入 $/M | 输出 $/M | 10$ 能跑 | AA 智能 | 国内 |
|---|---|---|---|---|---|
| inclusionAI: Ling-2.6-flash Inclusionai | $0.0100 | $0.0300 | 333.3M | — | 需代理 |
| Mistral: Mistral Nemo Mistralai | $0.0190 | $0.0300 | 333.3M | — | 需代理 |
| Sao10K: Llama 3 8B Lunaris | $0.0400 | $0.0500 | 200.0M | — | |
| MythoMax 13B | $0.0600 | $0.0600 | 166.7M | — | |
| Meta: Llama 3.1 8B Instruct Meta Llama | $0.0500 | $0.0800 | 125.0M | — | 需代理 |
| Mistral: Mistral Small 3 | $0.0500 | $0.0800 | 125.0M | — | |
| Google: Gemma 3 4B | $0.0500 | $0.10 | 100.0M | — | |
| IBM: Granite 4.1 8B | $0.0500 | $0.10 | 100.0M | — | |
| Mistral: Ministral 3 3B 2512 | $0.10 | $0.10 | 100.0M | — | |
| Nex AGI: Nex-N2-Mini | $0.0250 | $0.10 | 100.0M | — | |
| Qwen: Qwen2.5 7B Instruct | $0.0400 | $0.10 | 100.0M | — | |
| Reka Edge | $0.10 | $0.10 | 100.0M | — | |
| IBM: Granite 4.0 Micro Ibm Granite | $0.0170 | $0.11 | 89.3M | — | 需代理 |
| Google: Gemma 3n 4B | $0.0600 | $0.12 | 83.3M | — | |
| Poolside Laguna XS 2.1 Poolside | $0.0600 | $0.12 | 83.3M | — | 待确认 |
| OpenAI: gpt-oss-20b Openai | $0.0300 | $0.13 | 76.9M | — | 需代理 |
| Amazon: Nova Micro 1.0 | $0.0350 | $0.14 | 71.4M | — | |
| Microsoft: Phi 4 | $0.0700 | $0.14 | 71.4M | — | |
| Cohere: Command R7B (12-2024) | $0.0375 | $0.15 | 66.7M | — | |
| Google: Gemma 3 12B | $0.0500 | $0.15 | 66.7M | — | |
| Mistral: Ministral 3 8B 2512 | $0.15 | $0.15 | 66.7M | — | |
| Qwen: Qwen3.5-9B Qwen | $0.1000 | $0.15 | 66.7M | — | 需代理 |
| OpenAI: gpt-oss-120b Openai | $0.0370 | $0.17 | 58.8M | — | 需代理 |
| Meta: Llama Guard 4 12B | $0.18 | $0.18 | 55.6M | — | |
| DeepSeek V4 Flash DeepSeek | $0.0938 | $0.19 | 53.3M | — | 可用 |
| NVIDIA: Nemotron 3 Nano 30B A3B Nvidia | $0.0500 | $0.20 | 50.0M | — | 需代理 |
| ByteDance: UI-TARS 7B | $0.10 | $0.20 | 50.0M | — | |
| Mistral: Ministral 3 14B 2512 | $0.20 | $0.20 | 50.0M | — | |
| Poolside: Laguna S 2.1 | $0.10 | $0.20 | 50.0M | — | |
| Reka Flash 3 | $0.10 | $0.20 | 50.0M | — | |
| Meta: Llama 3.2 1B Instruct | $0.0270 | $0.20 | 49.8M | — | |
| Tencent Hy3 preview Tencent | $0.0630 | $0.21 | 47.6M | — | 可用 |
| Amazon: Nova Lite 1.0 | $0.0600 | $0.24 | 41.7M | — | |
| Qwen: Qwen3 14B | $0.12 | $0.24 | 41.7M | — | |
| Qwen: Qwen3.5-Flash | $0.0650 | $0.26 | 38.5M | — | |
| Qwen: Qwen3 Coder 30B A3B Instruct | $0.0700 | $0.27 | 37.0M | — | |
| Qwen: Qwen3 32B | $0.0800 | $0.28 | 35.7M | — | |
| MiMo-V2.5 Xiaomi | $0.14 | $0.28 | 35.7M | — | 可用 |
| ByteDance Seed: Seed 1.6 Flash | $0.0750 | $0.30 | 33.3M | — | |
| Google: Gemma 3 27B Google | $0.1000 | $0.30 | 33.3M | — | 需代理 |
| Llama 4 Scout Meta | $0.1000 | $0.30 | 33.3M | — | 需代理 |
| Mistral: Mistral Small 3.2 24B | $0.10 | $0.30 | 33.3M | — | |
| Mistral: Voxtral Small 24B 2507 | $0.10 | $0.30 | 33.3M | — | |
| OpenAI: gpt-oss-safeguard-20b | $0.0750 | $0.30 | 33.3M | — | |
| Qwen: Qwen3 30B A3B Instruct 2507 | $0.10 | $0.30 | 33.3M | — | |
| StepFun: Step 3.5 Flash | $0.10 | $0.30 | 33.3M | — | |
| Meta: Llama 3.2 3B Instruct | $0.0509 | $0.34 | 29.9M | — | |
| Google: Gemma 4 26B A4B Google | $0.0700 | $0.34 | 29.4M | — | 需代理 |
| Google: Gemma 4 31B Google | $0.12 | $0.37 | 27.0M | — | 需代理 |
| Google: Gemini 2.5 Flash Lite Google | $0.1000 | $0.40 | 25.0M | — | 需代理 |
| Meta: Llama 3.3 70B Instruct Meta Llama | $0.13 | $0.40 | 25.0M | — | 需代理 |
| Poolside: Laguna M.1 Poolside | $0.20 | $0.40 | 25.0M | — | 需代理 |
| Z.ai: GLM 4.7 Flash Z Ai | $0.0605 | $0.40 | 25.0M | — | 需代理 |
| ByteDance Seed: Seed-2.0-Mini | $0.10 | $0.40 | 25.0M | — | |
| Meta: Llama 3.1 70B Instruct | $0.40 | $0.40 | 25.0M | — | |
| Nous: Hermes 4 70B | $0.13 | $0.40 | 25.0M | — | |
| OpenAI: GPT-4.1 Nano | $0.10 | $0.40 | 25.0M | — | |
| OpenAI: GPT-5 Nano | $0.0500 | $0.40 | 25.0M | — | |
| Qwen2.5 72B Instruct | $0.36 | $0.40 | 25.0M | — | |
| TheDrummer: UnslopNemo 12B | $0.40 | $0.40 | 25.0M | — | |
| DeepSeek: DeepSeek V3.2 Exp Deepseek | $0.27 | $0.41 | 24.4M | — | 需代理 |
| Qwen: Qwen3 VL 32B Instruct | $0.10 | $0.42 | 24.0M | — | |
| NVIDIA Nemotron 3 Super NVIDIA | $0.0800 | $0.45 | 22.2M | — | 需代理 |
| Qwen: Qwen3 8B | $0.12 | $0.46 | 22.0M | — | |
| Qwen: Qwen3 VL 8B Instruct | $0.12 | $0.46 | 22.0M | — | |
| AllenAI: Olmo 3 32B Think | $0.15 | $0.50 | 20.0M | — | |
| TheDrummer: Cydonia 24B V4.1 | $0.30 | $0.50 | 20.0M | — | |
| TheDrummer: Rocinante 12B | $0.25 | $0.50 | 20.0M | — | |
| Qwen: Qwen3 30B A3B | $0.13 | $0.52 | 19.2M | — | |
| Qwen: Qwen3 VL 30B A3B Instruct | $0.13 | $0.52 | 19.2M | — | |
| Qwen: Qwen3 235B A22B Instruct 2507 | $0.0900 | $0.55 | 18.2M | — | |
| Mistral: Mistral Small 3.1 24B | $0.35 | $0.56 | 18.0M | — | |
| Tencent: Hunyuan A13B Instruct | $0.14 | $0.57 | 17.5M | — | |
| Tencent Hy3 Tencent | $0.14 | $0.58 | 17.2M | — | 直接可用 |
| Cohere: Command R (08-2024) | $0.15 | $0.60 | 16.7M | — | |
| Kwaipilot: KAT-Coder-Air V2.5 | $0.15 | $0.60 | 16.7M | — | |
| Mistral: Saba | $0.20 | $0.60 | 16.7M | — | |
| Mistral: Mistral Small 4 Mistralai | $0.15 | $0.60 | 16.7M | — | 需代理 |
| OpenAI: GPT-4o-mini Openai | $0.15 | $0.60 | 16.7M | — | 需代理 |
| OpenAI: GPT-4o-mini (2024-07-18) | $0.15 | $0.60 | 16.7M | — | |
| OpenAI: GPT-4o-mini Search Preview | $0.15 | $0.60 | 16.7M | — | |
| Upstage: Solar Pro 3 | $0.15 | $0.60 | 16.7M | — | |
| WizardLM-2 8x22B | $0.62 | $0.62 | 16.1M | — | |
| inclusionAI: Ling-2.6-1T | $0.0750 | $0.63 | 16.0M | — | |
| inclusionAI: Ring-2.6-1T | $0.0750 | $0.63 | 16.0M | — | |
| Google: Gemma 2 27B | $0.65 | $0.65 | 15.4M | — | |
| ReMM SLERP 13B | $0.45 | $0.65 | 15.4M | — | |
| Nous: Hermes 3 70B Instruct | $0.70 | $0.70 | 14.3M | — | |
| Inception: Mercury 2 | $0.25 | $0.75 | 13.3M | — | |
| Mancer: Weaver (alpha) | $0.50 | $0.75 | 13.3M | — | |
| Sao10K: Llama 3.3 Euryale 70B | $0.65 | $0.75 | 13.3M | — | |
| Qwen: Qwen-Plus | $0.26 | $0.78 | 12.8M | — | |
| Qwen: Qwen Plus 0728 | $0.26 | $0.78 | 12.8M | — | |
| Qwen: Qwen Plus 0728 (thinking) | $0.26 | $0.78 | 12.8M | — | |
| Qwen: Qwen3 Next 80B A3B Instruct | $0.0975 | $0.78 | 12.8M | — | |
| Qwen: Qwen3 Next 80B A3B Thinking | $0.0975 | $0.78 | 12.8M | — | |
| Llama 4 Maverick Meta | $0.20 | $0.80 | 12.5M | — | 需代理 |
| Arcee AI: Trinity Large Thinking | $0.25 | $0.80 | 12.5M | — | |
| DeepSeek: R1 Distill Llama 70B | $0.80 | $0.80 | 12.5M | — | |
| Qwen: Qwen3 Coder Next | $0.11 | $0.80 | 12.5M | — | |
| TheDrummer: Skyfall 36B V2 | $0.55 | $0.80 | 12.5M | — | |
| DeepSeek: DeepSeek V3 | $0.20 | $0.80 | 12.5M | — | |
| Sao10K: Llama 3.1 Euryale 70B v2.2 | $0.85 | $0.85 | 11.8M | — | |
| Z.ai: GLM 4.5 Air | $0.13 | $0.85 | 11.8M | — | |
| DeepSeek V4 Pro DeepSeek | $0.43 | $0.87 | 11.5M | — | 可用 |
| MiMo-V2.5-Pro Xiaomi | $0.43 | $0.87 | 11.5M | 42 | 可用 |
| MiniMax: MiniMax M2.5 Minimax | $0.15 | $0.90 | 11.1M | — | 需代理 |
| Venice: Uncensored | $0.20 | $0.90 | 11.1M | — | |
| Mistral: Codestral 2508 | $0.30 | $0.90 | 11.1M | — | |
| Z.ai: GLM 4.6V | $0.30 | $0.90 | 11.1M | — | |
| DeepSeek: DeepSeek V3.1 Deepseek | $0.25 | $0.95 | 10.5M | — | 需代理 |
| Qwen: Qwen3 Coder Flash | $0.20 | $0.97 | 10.3M | — | |
| DeepSeek: DeepSeek V3.1 Terminus Deepseek | $0.27 | $1.00 | 10.0M | — | 需代理 |
| MiniMax M2.7 MiniMax | $0.25 | $1.00 | 10.0M | — | 可用 |
| Nex-N2-Pro Nex AGI | $0.25 | $1.00 | 10.0M | — | 需代理 |
| Nous: Hermes 3 405B Instruct | $1.00 | $1.00 | 10.0M | — | |
| Perplexity: Sonar | $1.00 | $1.00 | 10.0M | — | |
| Qwen2.5 Coder 32B Instruct | $0.66 | $1.00 | 10.0M | — | |
| Qwen: Qwen2.5 VL 72B Instruct | $0.80 | $1.00 | 10.0M | — | |
| Qwen: Qwen3 Coder 480B A35B | $0.30 | $1.00 | 10.0M | — | |
| Qwen: Qwen3.5-35B-A3B Qwen | $0.14 | $1.00 | 10.0M | — | 需代理 |
| Qwen: Qwen3.6 35B A3B Qwen | $0.14 | $1.00 | 10.0M | — | 需代理 |
| MiniMax: MiniMax-01 | $0.20 | $1.10 | 9.1M | — | |
| DeepSeek: DeepSeek V3 0324 Deepseek | $0.27 | $1.12 | 8.9M | — | 需代理 |
| Qwen: Qwen3.6 Flash Qwen | $0.19 | $1.13 | 8.9M | — | 需代理 |
| Step 3.7 Flash 阶跃星辰 (StepFun) | $0.20 | $1.15 | 8.7M | — | 可用 |
| Arcee AI: Virtuoso Large | $0.75 | $1.20 | 8.3M | — | |
| Kwaipilot: KAT-Coder-Pro V2 | $0.30 | $1.20 | 8.3M | — | |
| Meituan: LongCat 2.0 | $0.30 | $1.20 | 8.3M | — | |
| MiniMax: MiniMax M2 | $0.30 | $1.20 | 8.3M | — | |
| MiniMax: MiniMax M2-her | $0.30 | $1.20 | 8.3M | — | |
| MiniMax: MiniMax M2.1 | $0.30 | $1.20 | 8.3M | — | |
| MiniMax M3 MiniMax | $0.30 | $1.20 | 8.3M | — | 可用 |
| Morph: Morph V3 Fast | $0.80 | $1.20 | 8.3M | — | |
| Anthropic: Claude 3 Haiku | $0.25 | $1.25 | 8.0M | — | |
| Baidu: ERNIE 4.5 VL 424B A47B | $0.42 | $1.25 | 8.0M | — | |
| Deep Cogito: Cogito v2.1 671B | $1.25 | $1.25 | 8.0M | — | |
| GPT-5.4 Nano OpenAI | $0.20 | $1.25 | 8.0M | — | 需代理 |
| Relace: Relace Apply 3 | $0.85 | $1.25 | 8.0M | — | |
| Qwen3.7 Plus Alibaba | $0.32 | $1.28 | 7.8M | — | 可用 |
| Qwen: Qwen3 VL 8B Thinking | $0.12 | $1.36 | 7.3M | — | |
| Aion-3.0 Mini AionLabs | $0.70 | $1.40 | 7.1M | — | 需代理 |
| Gemini 3.1 Flash Lite Google | $0.25 | $1.50 | 6.7M | — | 不稳定 |
| Gemini 3.1 Flash Lite Image Google | $0.25 | $1.50 | 6.7M | — | 不可用 |
| Google: Gemini 3.1 Flash Lite Preview Google | $0.25 | $1.50 | 6.7M | — | 需代理 |
| Mistral: Mistral Large 3 2512 | $0.50 | $1.50 | 6.7M | — | |
| OpenAI: GPT-3.5 Turbo | $0.50 | $1.50 | 6.7M | — | |
| Perceptron: Perceptron Mk1 | $0.15 | $1.50 | 6.7M | — | |
| Qwen: Qwen3 30B A3B Thinking 2507 | $0.13 | $1.56 | 6.4M | — | |
| Qwen: Qwen3 VL 30B A3B Thinking | $0.13 | $1.56 | 6.4M | — | |
| Qwen: Qwen3.5 Plus 2026-02-15 | $0.26 | $1.56 | 6.4M | — | |
| AionLabs: Aion-2.0 | $0.80 | $1.60 | 6.3M | — | |
| AionLabs: Aion-RP 1.0 (8B) | $0.80 | $1.60 | 6.3M | — | |
| OpenAI: GPT-4.1 Mini | $0.40 | $1.60 | 6.3M | — | |
| Z.ai: GLM 4.7 Z Ai | $0.40 | $1.75 | 5.7M | — | 需代理 |
| Qwen: Qwen3.5 Plus 2026-04-20 | $0.30 | $1.80 | 5.6M | — | |
| Z.ai: GLM 4.5V | $0.60 | $1.80 | 5.6M | — | |
| Qwen: Qwen3 235B A22B | $0.46 | $1.82 | 5.5M | — | |
| Morph: Morph V3 Large | $0.90 | $1.90 | 5.3M | — | |
| Qwen: Qwen3 VL 235B A22B Instruct | $0.21 | $1.90 | 5.3M | — | |
| Qwen3.6 Plus Alibaba (阿里云百炼) | $0.33 | $1.95 | 5.1M | 40 | 可用 |
| ByteDance Seed: Seed 1.6 | $0.25 | $2.00 | 5.0M | — | |
| ByteDance Seed: Seed-2.0-Lite | $0.25 | $2.00 | 5.0M | — | |
| Mistral: Devstral 2 2512 | $0.40 | $2.00 | 5.0M | — | |
| Mistral: Mistral Medium 3 | $0.40 | $2.00 | 5.0M | — | |
| Mistral: Mistral Medium 3.1 | $0.40 | $2.00 | 5.0M | — | |
| OpenAI: GPT-3.5 Turbo (older v0613) | $1.00 | $2.00 | 5.0M | — | |
| OpenAI: GPT-3.5 Turbo Instruct | $1.50 | $2.00 | 5.0M | — | |
| OpenAI: GPT-5 Image Mini | $2.50 | $2.00 | 5.0M | — | |
| OpenAI: GPT-5.1-Codex-Mini | $0.25 | $2.00 | 5.0M | — | |
| Grok Build 0.1 xAI | $1.00 | $2.00 | 5.0M | — | 需代理 |
| Qwen: Qwen3.5-122B-A10B | $0.26 | $2.08 | 4.8M | — | |
| DeepSeek: R1 0528 | $0.50 | $2.15 | 4.7M | — | |
| MiniMax: MiniMax M1 | $0.55 | $2.20 | 4.5M | — | |
| Z.ai: GLM 4.5 | $0.60 | $2.20 | 4.5M | — | |
| Qwen: Qwen3.5 397B A17B Qwen | $0.39 | $2.34 | 4.3M | — | 需代理 |
| OpenAI: GPT Audio Mini | $0.60 | $2.40 | 4.2M | — | |
| GLM 5.2 Z.ai (智谱) | $0.78 | $2.46 | 4.1M | — | 可用 |
| Amazon: Nova 2 Lite | $0.30 | $2.50 | 4.0M | — | |
| Google: Nano Banana (Gemini 2.5 Flash Image) | $0.30 | $2.50 | 4.0M | — | |
| Google: Gemini 3.5 Flash-Lite | $0.30 | $2.50 | 4.0M | — | |
| MoonshotAI: Kimi K2 0905 | $0.60 | $2.50 | 4.0M | — | |
| MoonshotAI: Kimi K2 Thinking | $0.60 | $2.50 | 4.0M | — | |
| Grok 4.20 xAI | $1.25 | $2.50 | 4.0M | — | 需代理 |
| Grok 4.20 Multi-Agent xAI | $1.25 | $2.50 | 4.0M | — | 需代理 |
| Grok 4.3 xAI | $1.25 | $2.50 | 4.0M | — | 需代理 |
| GLM 5 Z.ai (智谱) | $0.95 | $2.55 | 3.9M | — | 可用 |
| Qwen: Qwen3 VL 235B A22B Thinking | $0.26 | $2.60 | 3.8M | — | |
| Qwen: Qwen3.5-27B Qwen | $0.26 | $2.60 | 3.8M | — | 需代理 |
| Qwen: Qwen3.6 27B Qwen | $0.45 | $2.70 | 3.7M | — | 需代理 |
| MoonshotAI: Kimi K2.5 | $0.57 | $2.85 | 3.5M | — | |
| Kwaipilot: KAT-Coder-Pro V2.5 | $0.74 | $2.96 | 3.4M | — | |
| Gemini 3 Flash Preview Google | $0.50 | $3.00 | 3.3M | — | 不稳定 |
| Gemini 3.1 Flash Image (Nano Banana 2) Google | $0.50 | $3.00 | 3.3M | — | 需代理 |
| Google: Nano Banana 2 (Gemini 3.1 Flash Image Preview) | $0.50 | $3.00 | 3.3M | — | |
| Nous: Hermes 4 405B | $1.00 | $3.00 | 3.3M | — | |
| Qwen: Qwen3 235B A22B Thinking 2507 | $0.30 | $3.00 | 3.3M | — | |
| Relace: Relace Search | $1.00 | $3.00 | 3.3M | — | |
| GLM 5.1 Z.ai (智谱) | $0.97 | $3.04 | 3.3M | — | 可用 |
| Amazon: Nova Pro 1.0 | $0.80 | $3.20 | 3.1M | — | |
| Kimi K2.6 Moonshot AI | $0.68 | $3.42 | 2.9M | — | 可用 |
按场景四档推荐
第一档：极致便宜，跑量无感 — DeepSeek V4 Flash
输出 $0.197/M，10 美元能跑 5000 万 token。OpenRouter 周用量 3.53T 排全球第 1，开发者真金白银投出来的票。 国内直连，不用代理。智能分虽然不高（AA 中国榜第 10），但客服、摘要、分类、内容过滤这类「不需要天才，只需要不犯错」的场景完全够用。 同价位没有对手 — MiMo-V2.5 虽然降到 $0.28/M，但智能分 49 还不如 DeepSeek V4 Flash 的上下文（1M vs 1M 平手，但 DeepSeek 生态更成熟）。
适合：客服机器人、批量数据处理、RAG reader、个人日常聊天
第二档：便宜 + 能打 — MiMo-V2.5、MiniMax M2.7
MiMo-V2.5 2026-05-30 大幅降价：输入 $0.14、输出 $0.28/M，10 美元能跑 3500 万 token。 小米经济版，AA Intelligence 40，和 DeepSeek V4 Flash 智能相当，价格只贵 40%。国内直连，适合「便宜但要有点脑子」的场景。
MiniMax M3 输入 $0.279、输出 $1.2/M，10 美元能跑 830 万 token。 AA Intelligence 44 卡在国内 Top 10 门槛，Agent 能力比编程能力更强 — 更适合对话和轻量任务编排。 是国内 Top 10 里最便宜的输入价。
适合：中小规模对话系统、轻量 Agent、内容生成
第三档：性价比甜点 — GLM 5、GPT-5.4 Nano、Gemini 3 Flash
GLM 5 输出 $1.92/M，SuperCLUE 幻觉控制 86.85 全榜第一。开源可本地部署，国内节点稳定。 如果任务对「说瞎话」零容忍，GLM 5 是经济档里幻觉控制最好的。
GPT-5.4 Nano 输出 $1.25/M，Vectara HHEM 幻觉率 3.1% 全球第 2（仅次于蚂蚁 finix），速度 167 token/s。 需要代理，但如果你在海外或者有稳定中转，Nano 是 RAG 和摘要的隐形冠军。
Gemini 3 Flash Preview 输出 $3/M，速度 182 token/s 全场最快之一，1M 上下文。 国内访问不稳定，但速度和窗口在这个价位段没有对手。
适合：RAG 系统、严控幻觉的摘要、速度敏感场景
第四档：便宜旗舰 — Kimi K2.6（降价后）、DeepSeek V4 Pro
这两款的输出价仍属经济档（Kimi $3.42、V4 Pro $0.87），智能分 43-44 区间，属于「便宜版本的旗舰」。 Kimi K2.6 2026-05-30 降价后输出 $3.42/M，比 Claude Sonnet 4.6 便宜 4 倍，AA 中国 #5 智能。 DeepSeek V4 Pro 输出 $0.87/M，AA 中国 #4，开源可部署。 如果你的预算能到这个档位，能力会有质的飞跃 — 从「能用」变成「好用」。
适合：国内代码主力、复杂 Agent、需要智能和性价比平衡的场景
一句话结论
· 只图便宜 + 国内直连 → DeepSeek V4 Flash（$0.197/M，用量全球第 1）
· 便宜但不想太笨 → MiMo-V2.5（$0.28/M，小米大幅降价后性价比突出）
· 严控幻觉 + 国内 → GLM 5（开源 + 幻觉控制全榜第一）
· 严控幻觉 + 海外 → GPT-5.4 Nano（HHEM 3.1%，全球第 2）
· 速度优先 + 长上下文 → Gemini 3 Flash（182 t/s + 1M 窗口）
· 预算够到 $3/M 档 → 直接上 Kimi K2.6 或 DeepSeek V4 Pro，能力质变