模型对比
并排对比 System One 决策模型:每百万 token 输入价、免费输出、厂商与发布日期,附价格分层推荐(最便宜 / 均衡 / 旗舰)与一行换模型的代码。
价格矩阵(实时)
当前 35 个模型在售——数据由网关实时渲染。 下表在请求时实时拉取 GET /v1/models 渲染(绝非构建期快照);MCP 的 list_models 工具返回同源数据。
| 模型 | 厂商 | 输入 $/百万 | 档位 |
|---|---|---|---|
| baseten/joey-1 | Baseten | $0.01 | 最便宜 |
| perplexity/pplx-decider-flash-v1.2 | Perplexity | $0.01 | 最便宜 |
| respan/span-02 | Respan | $0.01 | 最便宜 |
| respan/span-01-lite | Respan | $0.01 | 最便宜 |
| parallel/decisions-fast | Parallel | $0.01 | 最便宜 |
| perplexity/pplx-decider-v1.3 | Perplexity | $0.01 | 最便宜 |
| perplexity/pplx-decider-flash-v1.3 | Perplexity | $0.01 | 最便宜 |
| liquid/d1 | Liquid AI | $0.02 | 均衡 |
| liquid/d1.1 | Liquid AI | $0.02 | 均衡 |
| perplexity/pplx-decider-v1.2 | Perplexity | $0.02 | 均衡 |
| respan/span-01 | Respan | $0.02 | 均衡 |
| liquid/d1.2 | Liquid AI | $0.02 | 均衡 |
| baseten/elephant-1 | Baseten | $0.035 | 均衡 |
| cloudflare/clef-flash | Cloudflare | $0.038 | 均衡 |
| nace-ai/drex-v1.5 | Nace.AI | $0.04 | 均衡 |
| scaledown/classify-1 | ScaledDown | $0.04 | 均衡 |
| liquid/d1-ultra | Liquid AI | $0.042 | 均衡 |
| microsoft/microsoft-decision-1 | Microsoft | $0.042 | 均衡 |
| togethercomputer/tev1-4b-experimental | Together | $0.042 | 均衡 |
| jaredpalmer/kev-4b | Jared Palmer | $0.042 | 均衡 |
| morph/reflex | Morph | $0.042 | 均衡 |
| typesafe/jev-1.13 | TypeSafe AI | $0.042 | 均衡 |
| typesafe/jev-1.14 | TypeSafe AI | $0.042 | 均衡 |
| typesafe/jev-latest | TypeSafe AI | $0.042 | 均衡 |
| liquid/d1.2-ultra | Liquid AI | $0.042 | 均衡 |
| upstage/solar-decide | Upstage | $0.05 | 旗舰 |
| upstage/solar-decide-flash | Upstage | $0.05 | 旗舰 |
| nace-ai/drex-max | Nace.AI | $0.06 | 旗舰 |
| nace-ai/drex-v2.0 | Nace.AI | $0.06 | 旗舰 |
| nace-ai/drex-max-finance | Nace.AI | $0.06 | 旗舰 |
| heygen/heygen-decision-1-turbo | heygen | $0.09 | 旗舰 |
| baseten/mammoth-1 | Baseten | $0.1 | 旗舰 |
| heygen/heygen-decision-1 | heygen | $0.14 | 旗舰 |
| cloudflare/clef-omni | Cloudflare | $0.15 | 旗舰 |
| cloudflare/clef | Cloudflare | $0.24 | 旗舰 |
按价格分层推荐
baseten/joey-1 · $0.01, perplexity/pplx-decider-flash-v1.2 · $0.01, respan/span-02 · $0.01, respan/span-01-lite · $0.01, parallel/decisions-fast · $0.01
liquid/d1 · $0.02, liquid/d1.1 · $0.02, perplexity/pplx-decider-v1.2 · $0.02, respan/span-01 · $0.02, liquid/d1.2 · $0.02
upstage/solar-decide · $0.05, upstage/solar-decide-flash · $0.05, nace-ai/drex-max · $0.06, nace-ai/drex-v2.0 · $0.06, nace-ai/drex-max-finance · $0.06
怎么选
所有模型共用同一请求结构,所以「你的任务上最便宜且够准」的就是对的:从最便宜档起步,用自己的样本评估准确率,只在低档误判处升档。因为换模型只是改一行 model,第一天起就能把不同流量分级路由到不同档位。
SWITCH
换模型只需改一行——整个目录的请求结构完全一致,可以按请求做模型 A/B 对比。