Inception AI Mercury
Listed at https://inceptionlabs.ai
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Mercury 2.5 (Preview): $0.20/M (cached $0.02/M, output $0.75/M); Mercury 2 / Mercury Edit 2: $0.25/M input (cached $0.025/M) + $0.75/M output | per million tokens |
| Output | All Mercury models: $0.75/M output at list; Mercury 2.5 Preview currently discounted on OpenRouter at ~$0.04/M in / $0.15/M out (launch promo, context 260K) | per million tokens |
🤖 Supported Models (3)
✨ Pros
- ✓World's first commercially available family of diffusion LLMs (dLLMs) that generate tokens in parallel
- ✓Mercury 2 runs 1,009+ tokens/sec on NVIDIA Blackwell with sub-300ms time-to-first-token
- ✓Mercury 2.5 Preview (on OpenRouter since 2026-08-31) — the smartest reasoning dLLM, 260K context
- ✓OpenAI-compatible /v1 API plus official Python + TypeScript SDKs for zero-friction migration
- ✓100M-token free tier and sub-frontier pricing ($0.25/$0.75 per 1M) vs auto-regressive frontier models
⚠️ Cons
- ×Mercury 2.5 pricing appears only on the marketing card — the official docs price table still lists only Mercury 2 / Edit 2
- ×Two coexisting price sets can confuse: OpenRouter launch promo (~$0.04/$0.15) vs vendor card list ($0.20/$0.75)
- ×Text-only models — no vision or multimodal input on Mercury 2 / Edit 2
- ×No mainland-China edge; mainland developers need a proxy / domestic fallback
- ×Mercury 2.5 is a Preview — final pricing and capabilities may change
🎯 Best For
Latency- and throughput-sensitive agentic, real-time autocomplete, and drafting workloads that benefit from parallel diffusion decoding; teams wanting frontier-class reasoning and code at sub-frontier per-million pricing; developers exploring the diffusion-LLM paradigm.
💰 Pricing & Plans
| Model | Input ($/M) | Output ($/M) | Notes |
|---|---|---|---|
| Mercury 2.5 (Preview) | $0.20 ($0.02 cached) | $0.75 | 260K ctx; flagship reasoning dLLM. OpenRouter launch promo ~$0.04/$0.15 |
| Mercury 2 | $0.25 ($0.025 cached) | $0.75 | 128K ctx; ~1,009 tok/s; tool calling + structured outputs |
| Mercury Edit 2 | $0.25 ($0.025 cached) | $0.75 | 32K ctx; code FIM & NextEdit completion |
🔧 API & Developer Experience
- •OpenAI-Compatible Base URL: Drop-in OpenAI-compatible endpoint at https://api.inceptionlabs.ai/v1 (chat at /v1/chat/completions, plus /v1/fim/completions and /v1/edit/completions for code). Existing OpenAI SDKs work by swapping the base URL and API key.
- •Official SDKs: First-party 'inceptionai' client libraries for Python (pip install inceptionai) and TypeScript/JavaScript (npm install inceptionai); usage is from inceptionai import Inception with the INCEPTION_API_KEY env var.
- •Diffusion Streaming & Denoising: Streaming emits refined 'blocks' of text rather than one token at a time; a diffusing:true parameter lets you visualise the coarse-to-fine denoising — the signature of a diffusion decoder.
- •Tool Calling + Structured Outputs: Native function/tool calling and schema-aligned JSON output on Mercury 2 and Mercury 2.5 — enough for agentic and retrieval pipelines without glue.
- •Free Tier & Rate Limits: Every new account gets 100M free tokens with no payment details. Free tier caps at 1,000 req/min and 1M input tokens/min; Pay-As-You-Go raises this to 3,000 req/min and 3M input tokens.
- •Tunable Reasoning: Mercury 2 introduced tunable reasoning effort, letting you trade compute for answer quality — useful for controlling cost on high-volume QA.
⚡ Why Diffusion: Parallel Token Generation
Inception's Mercury is the first commercially available family of diffusion LLMs, and it changes the decoding economics. Traditional auto-regressive LLMs commit to text one token at a time — a sequential bottleneck. Mercury instead denoises its output: it starts from noise over a target-length span and iteratively refines whole blocks of text toward the final answer in a handful of steps, generating many tokens in parallel. On the same NVIDIA Blackwell GPUs that run the auto-regressive frontier, Mercury 2 sustains 1,009+ tokens/sec with a sub-300ms time-to-first-token and 5-7x higher throughput, at up to 70% lower cost per task. For code completion and interactive agents, latency and throughput matter as much as benchmark quality, which is where diffusion's coarse-to-fine decode shows up in practice. The chief trade-off to weigh is that Mercury is text-only today and Mercury 2.5 currently ships through an OpenRouter preview rather than the full native API.
🌐 Regional Availability & Latency
Inception AI is headquartered in Redwood City, California, and Mercury's hosted inference API (api.inceptionlabs.ai) is served from US infrastructure. North America and Europe see the lowest latency, with the diffusion decoder's parallel decode making interactive chat and code completion feel faster than the per-token TTFT of an equivalent auto-regressive model. Asia-Pacific developers reach the US origin over transpacific fiber and can expect higher round-trip times; as with most US-hosted LLM APIs this is fine for batch and agentic jobs that tolerate a slower first byte, and noticeable for ultra-low-latency streaming. There is no mainland-China edge and no Chinese payment or data-residency option, so mainland developers should route traffic through a proxy or use a domestic provider (Qwen, Zhipu, Doubao) for any user-facing or latency- and residency-sensitive China surface.