可进入对比
档案摘要 / Decision snapshot
先判断能不能用于选型。
有真实任务证据
来自模型卡
边界优先
30 秒结论
可进入选型对比,但仍以案例证据为准。
来自公开模型资料库,并已补厂商/官方/模型家族证据;当前站内暂无可核验 A 类案例。
这页有来源、有风险说明、有公开可核验 A 类案例。
数据可信度 / 证据强度
强
A 类案例
7
AA 评分
33.93
官方来源
3 个补充入口
案例来源
7 条可核验
来自公开资料与模型数据库 + Artificial Analysis + 3 个厂商/官方/模型家族证据入口;AA 不是唯一事实来源。
Evidence Distribution
案例证据分布
DeepSeek V3.1 Terminus 当前关联 7 条 A 类案例;这里按任务、来源和复核时间观察证据结构。
Evidence Snapshot
证据快照覆盖
DeepSeek V3.1 Terminus 的 7 条 A 类案例中,已有 7 条完成本地证据快照;0 条仍在快照队列。
可用 manifest 复核
等待落盘
需补抓
需人工复核
证据完整度
5 / 5
Feishu Bitable
3 个入口
7 条
已绑定
已标注
能力边界
用标签表达信号,不用标签替代证据。
通用/待核验
暂无数据
待核验
开放/混合信号
中
| 厂商 | DeepSeek |
|---|---|
| 厂商页 | DeepSeek |
| 发布时间 | 2025-09-22 |
| 模型 ID | deepseek-v3-1-terminus |
| 上下文 | 暂无数据 |
| 输出 | 暂无数据 |
| 模态 | 文本 / 多模态待核验 |
| 推理 | 是 / 推理模型或 thinking 模式 |
| 价格 | 官方未披露 / 暂无数据 |
| 可用平台 | DeepSeek |
| AA 评分 | 33.93 |
| 基础模型发布时间 | Artificial Analysis 未披露 |
| 资料来源 | 公开资料、厂商信息和案例库 |
| A 类案例 | 7 |
谱系位置
它在厂商路线中的位置
此页将 DeepSeek V3.1 Terminus 放入 DeepSeek 的当前 Atlas 路线中。若同厂前后代资料不足,先保留为路线追踪入口,不脑补谱系关系。
2025-09-22
DeepSeek V3.1 Terminus 资料状态
来自公开资料与模型数据库 + Artificial Analysis + 3 个厂商/官方/模型家族证据入口;AA 不是唯一事实来源。
公开档案
模型状态
DeepSeek V3.1 Terminus 当前标记为已有真实案例。
案例库
公开案例复核
已有 7 条可核验 A 类案例,达到完整补齐线
案例时间线
最近进入档案的 A 类案例
用采集时间呈现证据进入 Atlas 的顺序,方便复核来源新鲜度。
2026-06-27T09:42:45Z
QuantumNous/new-api 用户配置 DeepSeek V3.1 Terminus 的 SiliconFlow 与 NVIDIA 双渠道并暴露大小写匹配问题
zhanwuzhijing / QuantumNous new-api / api_gateway_integration
2026-06-27T09:38:41Z
sammy22cool built an OpenAI-compatible proxy that routes requests to DeepSeek V3.1 Terminus via NVIDIA NIM
sammy22cool / api_integration
2026-06-27T09:38:41Z
com-wuqi published a FastAPI user-management and authentication service built with DeepSeek V3.1 Terminus through SiliconFlow
com-wuqi / software_development
2026-06-27T06:03:28Z
rick-stevens-ai 在 8 张 Intel Data Center GPU Max 1550 上运行 DeepSeek-V3.1-Terminus
rick-stevens-ai / inference_infrastructure
2026-06-27T06:03:28Z
RikkaHub 用户验证 DeepSeek-V3.1-Terminus 在 SiliconFlow/DeepSeek API 下的 thinking 参数支持
Zhichii / RikkaHub / app_integration
适合场景
- - 作为完整模型索引和厂商路线追踪入口
- - 推理/agentic workflow 候选
不适合
- - 缺少真实案例时不要包装成推荐
- - 缺字段时不要脑补价格、性能或上下文
真实案例
可核验 A 类案例精选
当前展示 7 条代表案例;排序优先看模型卡精选、证据可信度、快照状态和公开仓库/技术任务信号。案例总数仍为 7。
Representative Evidence
代表案例排序
当前第一代表案例是“QuantumNous/new-api 用户配置 DeepSeek V3.1 Terminus 的 SiliconFlow 与 NVIDIA 双渠道并暴露大小写匹配问题”。它会优先出现在本页,是因为该案例同时进入模型卡精选、具备 100/100 的证据可信度,并带有 已快照证据。
模型卡精选优先
A+ 完整链路
archive/evidence/deepseek-v3-1-terminus-quantumnous-new-api-channel-case-matching/manifest.json
Coding / repo / agent
zhanwuzhijing / QuantumNous new-api 使用 DeepSeek V3.1 Terminus 处理代码审查和测试生成
zhanwuzhijing / QuantumNous new-api · DeepSeek V3.1 Terminus
A+ 完整链路 · 代码仓库证据
zhanwuzhijing / QuantumNous new-api 公开的代码审查与测试案例,来源为 公开代码库,复核于 2026-06-27T09:42:45Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。
任务:这是一个围绕代码审查和测试生成的真实任务,公开材料可以回溯到具体使用者和具体产物。 原始资料写作:A QuantumNous/new-api user configured DeepSeek V3.1 Terminus through two upstream providers, SiliconFlow using deepseek-ai/DeepSeek-V3.1-Terminus and NVIDIA using deepseek-ai/deepseek-v3.1-terminus, then used the new-ap…
公开产物:公开材料提供公开代码、README 或项目配置,可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作:The public issue documents a reproducible integration artifact with screenshots: new-api treats the upper-case and lower-case provider model identifiers as the same model for duplicate checks, but the bound-channel disp…
模型作用:DeepSeek V3.1 Terminus 在该案例中承担代码审查和测试生成相关的生成、分析、编排或实现角色。 原始资料写作:DeepSeek V3.1 Terminus is the concrete model being exposed through the gateway; its provider-specific identifiers drive the channel binding, routing, and model-management behavior that the user is testing in the product.
A 类理由:有具体使用者、具体任务、公开原始证据和可访问产物,可公开核验。
风险边界:当前判断基于公开材料;若产物下线、仓库变更或模型参与比例仅来自作者自述,需要在引用前重新复核。 原始资料写作:Evidence is a public GitHub issue with user-supplied configuration and screenshots rather than an official customer story; it is still a real integration/debugging use case with exact model IDs, named user, task, result…
原始记录:QuantumNous/new-api 用户配置 DeepSeek V3.1 Terminus 的 SiliconFlow 与 NVIDIA 双渠道并暴露大小写匹配问题
com-wuqi 使用 DeepSeek V3.1 Terminus 处理软件工程任务执行
com-wuqi · DeepSeek V3.1 Terminus
A+ 完整链路 · 代码仓库证据
com-wuqi 公开的代码代理与软件工程案例,来源为 公开代码库,复核于 2026-06-27T09:38:41Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。
任务:这是一个围绕软件工程任务执行的真实任务,公开材料可以回溯到具体使用者和具体产物。 原始资料写作:Generate and publish a FastAPI, SQLite, SQLModel, JWT, and bcrypt based user registration, login, current-user, and admin user-management backend service.
公开产物:公开材料提供公开代码、README 或项目配置,可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作:The public repository contains the completed FastAPI authentication server with app modules, route files, startup script, API test script, dependency list, README, and documented endpoints including /auth/register, /aut…
模型作用:DeepSeek V3.1 Terminus 在该案例中承担软件工程任务执行相关的生成、分析、编排或实现角色。 原始资料写作:The repository description explicitly says the project was completed by Claude Code with SiliconFlow and DeepSeek-V3.1-Terminus; the model is therefore tied to code-generation and implementation assistance for the backe…
A 类理由:有具体使用者、具体任务、公开原始证据和可访问产物,可公开核验。
风险边界:当前判断基于公开材料;若产物下线、仓库变更或模型参与比例仅来自作者自述,需要在引用前重新复核。 原始资料写作:Evidence is a GitHub repository description plus the published code artifact; model usage is self-reported by the repository owner and includes Claude Code/SiliconFlow in the toolchain, so contribution is assistant-assi…
原始记录:com-wuqi published a FastAPI user-management and authentication service built with DeepSeek V3.1 Terminus through SiliconFlow
sammy22cool 使用 DeepSeek V3.1 Terminus 处理软件工程任务执行
sammy22cool · DeepSeek V3.1 Terminus
A+ 完整链路 · 代码仓库证据
sammy22cool 公开的代码代理与软件工程案例,来源为 公开代码库,复核于 2026-06-27T09:38:41Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。
任务:这是一个围绕软件工程任务执行的真实任务,公开材料可以回溯到具体使用者和具体产物。 原始资料写作:Create a Node.js/Express proxy exposing OpenAI-compatible /v1/models and /v1/chat/completions endpoints while routing incoming chat completion requests to the NVIDIA NIM model id deepseek-ai/deepseek-v3.1-terminus.
公开产物:公开材料提供公开代码、README 或项目配置,可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作:The public repository contains a runnable Express service, package.json, health check, model-list endpoint, and chat-completions handler; server.js maps common OpenAI, Claude, Gemini, and DeepSeek model aliases to deeps…
模型作用:DeepSeek V3.1 Terminus 在该案例中承担软件工程任务执行相关的生成、分析、编排或实现角色。 原始资料写作:DeepSeek V3.1 Terminus is the backend inference model for the proxy; all compatible client requests are normalized and routed to the Terminus model so downstream tools can use it through an OpenAI-style API surface.
A 类理由:有具体使用者、具体任务、公开原始证据和可访问产物,可公开核验。
风险边界:当前判断基于公开材料;若产物下线、仓库变更或模型参与比例仅来自作者自述,需要在引用前重新复核。 原始资料写作:Evidence is a public GitHub repository and source file, not an official customer story; the implementation is self-published by the repository owner but the exact model id is present in code and the artifact is reachabl…
原始记录:sammy22cool built an OpenAI-compatible proxy that routes requests to DeepSeek V3.1 Terminus via NVIDIA NIM
rick-stevens-ai 使用 DeepSeek V3.1 Terminus 处理智能体流程编排
rick-stevens-ai · DeepSeek V3.1 Terminus
A+ 完整链路 · 代码仓库证据
rick-stevens-ai 公开的智能体工作流案例,来源为 公开代码库,复核于 2026-06-27T06:03:28Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。
任务:公开仓库提供在 Intel Data Center GPU Max 1550 上用 llama.cpp SYCL 后端和多 GPU layer splitting 运行 DeepSeek-V3.1-Terminus 671B/405GB Q4_K_M 的完整配置。
公开产物:仓库描述给出实际运行结果:8 张 GPU 上实现 OpenAI-compatible API server,并报告 4.1 tok/s generation across 8 GPUs;README 列出 131K context、9 个 GGUF 分片、硬件/软件要求和启动配置。
模型作用:DeepSeek-V3.1-Terminus 是被部署和服务的核心大模型,用于代码生成、推理和 agent 任务;该项目验证了其在 Intel MAX GPU 多卡环境中的推理可行性。
A 类理由:有具体使用者、具体任务、公开原始证据和可访问产物,可公开核验。
风险边界:这是工程部署/推理基础设施案例,不是终端业务 customer story;但仓库明确绑定 exact model、组织、任务和可访问产物。
原始记录:rick-stevens-ai 在 8 张 Intel Data Center GPU Max 1550 上运行 DeepSeek-V3.1-Terminus
Zhichii / RikkaHub 使用 DeepSeek V3.1 Terminus 处理真实任务执行
Zhichii / RikkaHub · DeepSeek V3.1 Terminus
A+ 完整链路 · 代码仓库证据
Zhichii / RikkaHub 公开的真实任务执行案例,来源为 公开代码库,复核于 2026-06-27T06:03:28Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。
任务:RikkaHub 用户在 issue 中列出 deepseek-ai/DeepSeek-V3.1-Terminus 和 Pro/deepseek-ai/DeepSeek-V3.1-Terminus 支持 enable_thinking,并在 Android 版 RikkaHub 1.6.16 中测试 Thinking 按钮/API 格式兼容性。
公开产物:Issue 记录了可复现的问题和 workaround:手动添加 Claude/Anthropic 格式的 SiliconFlow 或 DeepSeek API 后 thinking 参数可用;页面还附有展示硅基流动和 DeepSeek 方案的 mp4 截图证据。
模型作用:DeepSeek-V3.1-Terminus 是被集成到移动端聊天客户端的 thinking-capable 模型,用于产生带 thinking/reasoning 行为的对话输出;案例验证了其 API 参数在客户端中的实际适配路径。
A 类理由:有具体使用者、具体任务、公开原始证据和可访问产物,可公开核验。
风险边界:证据来自 GitHub issue 和附件,属于真实用户集成/调试场景;不是正式 customer story,但包含具体用户、模型、任务、结果和公开 artifact。
原始记录:RikkaHub 用户验证 DeepSeek-V3.1-Terminus 在 SiliconFlow/DeepSeek API 下的 thinking 参数支持
AIR-hl (research group) 使用 DeepSeek V3.1 Terminus 处理多模态内容处理
AIR-hl (research group) · DeepSeek V3.1 Terminus
A+ 完整链路 · 代码仓库证据
AIR-hl (research group) 公开的多模态生成与理解案例,来源为 公开代码库,复核于 2026-06-27T01:40:00Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。
任务:这是一个围绕多模态内容处理的真实任务,公开材料可以回溯到具体使用者和具体产物。 原始资料写作:Training DAPO (Direct Alignment from Preference Optimization) on DeepSeek-V3.1-Terminus with FP8 format for reinforcement learning to improve reasoning capabilities. The training runs on a 24-node cluster with 8 NVIDIA …
公开产物:公开材料提供公开代码、README 或项目配置,可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作:Encountered a dimension mismatch error (576 must be a multiple of block_size0: 128) during FP8 weight quantization in the verl training pipeline. The issue was filed on 2026-01-07 and received 14 comments, indicating ac…
模型作用:DeepSeek V3.1 Terminus 在该案例中承担多模态内容处理相关的生成、分析、编排或实现角色。 原始资料写作:DeepSeek-V3.1-Terminus serves as the base model for RL training. Its 671B-parameter MoE architecture with Mixture-of-Experts design enables efficient training at scale. The model's improved language consistency and agen…
A 类理由:有具体使用者、具体任务、公开原始证据和可访问产物,可公开核验。
风险边界:当前判断基于公开材料;若产物下线、仓库变更或模型参与比例仅来自作者自述,需要在引用前重新复核。 原始资料写作:The user encountered a technical error during FP8 training, indicating potential compatibility issues with certain quantization formats. The issue is actively being debugged. The model's large size (671B params) require…
原始记录:AIR-hl RL training with DAPO on DeepSeek-V3.1-Terminus using FP8 quantization
April-99 (developer) 使用 DeepSeek V3.1 Terminus 处理代码审查和测试生成
April-99 (developer) · DeepSeek V3.1 Terminus
A+ 完整链路 · 代码仓库证据
April-99 (developer) 公开的代码审查与测试案例,来源为 公开代码库,复核于 2026-06-27T01:40:00Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。
任务:这是一个围绕代码审查和测试生成的真实任务,公开材料可以回溯到具体使用者和具体产物。 原始资料写作:Deploying the quantized DeepSeek-V3.1-Terminus-w4a8-mtp-QuaRot model on Huawei Ascend NPUs using vllm-ascend 0.18.0rc1 inference engine. The deployment targets the w4a8 (4-bit weights, 8-bit activations) quantized varia…
公开产物:公开材料提供公开代码、README 或项目配置,可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作:Reported a bug where the model intermittently returns empty content in responses when sending specific request payloads. The issue was filed on the vllm-ascend project tracker, indicating active deployment and testing o…
模型作用:DeepSeek V3.1 Terminus 在该案例中承担代码审查和测试生成相关的生成、分析、编排或实现角色。 原始资料写作:DeepSeek-V3.1-Terminus is the core inference model being deployed. Its w4a8 quantized variant with QuaRot optimization enables efficient deployment on Ascend NPUs while maintaining model quality. The MTP (Mixture-of-Exp…
A 类理由:有具体使用者、具体任务、公开原始证据和可访问产物,可公开核验。
风险边界:当前判断基于公开材料;若产物下线、仓库变更或模型参与比例仅来自作者自述,需要在引用前重新复核。 原始资料写作:The deployment encountered intermittent empty response issues, suggesting the quantized model may have edge-case stability problems on Ascend NPU hardware. The vllm-ascend 0.18.0rc1 version is a release candidate, not a…
原始记录:April-99 deploying DeepSeek-V3.1-Terminus on Huawei Ascend NPUs with vllm-ascend
数据缺口
只基于现有字段判断缺口;缺失项不会被猜测填充。
- - 补充方向:核验 价格 的官方/API/案例来源。
- - 补充方向:核验 上下文 的官方/API/案例来源。
- - 补充方向:核验 输出 的官方/API/案例来源。
下一步验证建议
- - 继续保留 A 类案例的原始证据、产物页和版本快照。
- - 补官方发布/API/System Card 来源;若官方未披露,继续标注“官方未披露”。
- - 把 benchmark、教程、发布文保留为背景资料,不提升为真实案例。