Mercury dLLMs：Inception Labs的基于扩散的代码生成API

基准评分

Inception Labs Mercury dLLMs 在 Agent 就绪度与 AI 可见性上的得分 AI 就绪度和 GEO Score 是 VibeLaunch 在提交后生成的平台评估。

决策摘要

Software developers and engineering teams

概述

Inception Labs 推出了 Mercury dLLMs，这是大语言模型技术的一次革命性飞跃，旨在以显著降低的成本提供极速推理和前沿级质量。传统的 LLM 采用顺序生成文本的方式，一次生成一个 Token，这往往成为速度和效率的瓶颈。然而，Mercury 的扩散大语言模型 (dLLMs) 采用并行方式生成 Token，极大地提升了处理速度并最大化了 GPU 利用率。这种创新方法使其成为驱动新一代高要求 AI 应用的理想选择。\n\nMercury Diffusion Models 旨在克服传统 LLM 的局限性。通过实现并行文本生成，它们在性能上提供了实质性优势，为寻求集成尖端 AI 的企业提供了极具成本效益的解决方案。无论您是需要加速代码编写、实现实时语音交互、增强创意工作流，还是优化企业搜索，Mercury dLLMs 都能提供所需的响应速度和质量。\n\n### 核心能力\n- 并行 Token 生成：与顺序生成的 LLM 不同，Mercury dLLMs 同时生成 Token，从而显著提升速度并增强 GPU 效率。\n- 高质量输出：在 AI 生成内容方面达到前沿级质量，确保各种应用场景下结果的可靠性与专业性。\n- 成本效益：与传统模型相比，单位 Token 成本更低，使先进 AI 变得更易获取且更具经济可行性。\n- 128K 上下文窗口：凭借超大上下文窗口处理海量信息，支持更复杂、更细致的 AI 任务。\n\n### 驱动尖端 AI 应用\nMercury dLLMs 功能强大，可集成到广泛的应用中：\n- 极速代码编辑：体验响应迅速的自动补全和智能建议。\n- 实时语音智能体：在客户支持或翻译场景中进行自然、流畅的对话。\n- 快速创意副驾驶：缩短等待时间，加速编辑和创意工作。\n- 快速企业搜索：从庞大的知识库中即时检索相关数据。\n- 无缝企业工作流：利用超高响应速度的 AI 自动化复杂流程。\n\nInception Labs 还提供专门针对编程优化的 Mercury Coder，以及用于超低延迟应用的通用型 dLLM。两款模型均支持流式传输、工具调用（tool use）和结构化输出。针对企业需求，Inception Labs 通过 AWS Bedrock 等主流云供应商提供集成服务，并提供微调、私有化部署和专属支持选项。其模型兼容 OpenAI API，确保可以无缝替换现有的 LLM 集成方案。

评价 (0)

还没有评价。成为第一个评价的人！

评分构成

编辑评分由哪些维度构成，每项附判断依据。 AI 就绪度和 GEO Score 是 VibeLaunch 在提交后生成的平台评估。

Information quality

Only a small fraction of L1 passages contain substantive product information; the majority are Framer/CSS styling fragments. Key claims about speed and quality are vendor-provided with no methodology disclosure.

3.0

建议核验

Of approximately 45 L1 passage clusters, roughly 10 contain meaningful product information. Speed and quality claims lack benchmark methodology.

Ease of use

Standard API design with billing plans, key authentication, streaming, and familiar LLM parameters. No SDK or quickstart evidence, and pricing opacity complicates onboarding.

5.0

建议核验

API requires billing plan signup and key generation; supports streaming and standard parameters.

Feature depth

Two endpoints (completions and FIM), multi-language support, and a distinctive diffusion architecture. Limited to one model variant and one modality (code generation).

4.0

建议核验

FIM completions endpoint, 10+ language support, but only Mercury Coder Small documented.

Workflow fit

FIM is purpose-built for IDE integration. Sub-agent orchestration pattern is described but not backed by Mercury-specific case studies or integration documentation.

4.5

建议核验

FIM endpoint design aligns with code editor workflows; sub-agent pattern described in blog content.

Reliability

Terms explicitly permit service suspension or discontinuation. No SLA, uptime guarantee, or deprecation policy documented.

2.5

建议核验

Terms state services may be suspended or discontinued at any time with notice.

Value

No pricing information available in the source packet. Users cannot evaluate cost relative to alternatives.

2.0

建议核验

Billing plans are referenced but no specific pricing tiers or rates are published in available sources.

评分反映可查证的产品资料，不代表实际使用效果保证。

Agent 就绪度

评估 Agent 能否通过产品的官方信息理解产品，并重建一条有文档依据的工作流程。

Automated agent-readiness assessment of https://inceptionlabs.ai/: 9 of 22 checks verified across 2 fetched pages. No substantial machine interface is documented — agents can understand and cite the product but not operate it. Absent: agent_tooling_artifacts, request_examples, response_examples, version_information, changelog, cli.

就绪度维度

评估维度	得分
文档质量	85
执行结果可验证性	20
机器接口	35
项目定位清晰度	50
资源可发现性	100
工作流完整度	65

对 Agent 有帮助的部分

docs: verified during this run
llms txt: verified during this run
sitemap: verified during this run
quickstart: verified during this run
api reference: verified during this run
authentication: verified during this run

Agent 受阻的部分

No agent instruction files, code-distribution commands, or named slash-command skills found across fetched pages.
No request examples signal matched across 2 fetched pages.
No response examples signal matched across 2 fetched pages.
No version information signal matched across 2 fetched pages.
No changelog signal matched across 2 fetched pages.
No cli signal matched across 2 fetched pages.

检查项	状态	详情
理解产品3/5 已核验
产品文档	已核验	Developer/documentation pages reachable from the entry page (e.g. https://docs.inceptionlabs.ai/get-started/get-started).
快速开始	已核验	Probe matched on https://docs.inceptionlabs.ai/get-started/get-started: /quick ?start\|getting started\|in (five\|5)/.
API 参考	已核验	Probe matched on https://docs.inceptionlabs.ai/get-started/get-started: /api (reference\|documentation\|endpoints?)/.
请求示例	未在本次官方来源链中找到
响应示例	未在本次官方来源链中找到
连接接口2/4 已核验
SDK	已核验	Probe matched on https://docs.inceptionlabs.ai/get-started/get-started: /\bsdk\b\|client library\|npm package\|pip i/.
MCP 接口	未在本次官方来源链中找到
Webhooks	未在本次官方来源链中找到
认证文档	已核验	Probe matched on https://docs.inceptionlabs.ai/get-started/get-started: /api key\|bearer\|oauth\|access token\|authen/.
执行工作流0/6 已核验
命令行工具	未在本次官方来源链中找到
非交互式命令	不适用于该产品	No CLI was found to evaluate for this property.
命令行结构化输出	不适用于该产品	No CLI was found to evaluate for this property.
结构化导入与导出	未在本次官方来源链中找到
成功状态验证	未在本次官方来源链中找到
智能体工具产物	未在本次官方来源链中找到
维护与排错2/4 已核验
错误文档	已核验	Probe matched on https://docs.inceptionlabs.ai/get-started/get-started: /error (codes?\|handling\|responses?)\|4xx\|5/.
速率限制	已核验	Probe matched on https://docs.inceptionlabs.ai/get-started/get-started: /rate limit\|429\|throttl\|requests per (sec/.
版本信息	未在本次官方来源链中找到
更新日志	未在本次官方来源链中找到
发现与验证2/3 已核验
llms.txt	已核验	llms.txt published at the site root (64 lines).
站点地图	已核验	sitemap.xml reachable and lists site pages.
智能体原生定位	未在本次官方来源链中找到

证据核查

关于该工具的公开声明，每条均标注核验状态与引用来源。

blog/introducing-inception-api已验证6

www.inceptionlabs.ai已验证核验于 2026年7月16日

Mercury models are the first commercial-scale diffusion large language models (dLLMs).

Mercury Coder Small runs more than 5x faster than GPT-4o Mini and Claude 3.5 Haiku.

Mercury Coder Small matches GPT-4o Mini and Claude 3.5 Haiku in code generation quality.

The Inception API provides programmatic access to Mercury dLLMs with billing plans and API key authentication.

The API supports fill-in-the-middle (FIM) completions, where the model generates code between a prefix and suffix.

The API supports Python, JavaScript, Java, TypeScript, Bash, SQL, C, C++, PHP, HTML, and more languages.

https://www.inceptionlabs.ai/blog/introducing-inception-api

blog/mercury-2-the-first-reasoning-model-fast-enough-to-pick-up-the-phone厂商声明2

www.inceptionlabs.ai厂商声明核验于 2026年7月16日

Mercury models are the first commercial-scale diffusion large language models (dLLMs).

dLLMs use a diffusion architecture that generates tokens in parallel rather than sequentially.

https://www.inceptionlabs.ai/blog/mercury-2-the-first-reasoning-model-fast-enough-to-pick-up-the-phone

docs/terms-of-use已验证2

www.inceptionlabs.ai已验证核验于 2026年7月16日

Inception Labs may suspend, discontinue, or modify services at any time with notice.

https://www.inceptionlabs.ai/docs/terms-of-use

Inception – When Every Millisecond Matters已验证1