Radar / AI models / Mercury 3

Mercury 3: Inception's 1M-context diffusion LLM at 1,068 tokens per second

Inception released Mercury 3 on Oct 19, a diffusion language model that drafts many tokens in parallel instead of one at a time, reaching 1,068 tokens per second on NVIDIA GPUs with a 1M-token context window, tool calling and structured outputs, priced at $0.20/$0.70 per million input/output tokens (50% off at launch). Inception says it matches or beats GPT-6 Luna and Claude Haiku 5.5 on its own benchmarks.

new launchmodels · frontierJEV traction 0.47added 2026-10-24

Open Mercury 3 →View on the radar

Why it matters

The first diffusion LLM claimed to match speed-tier frontier models on quality while running 4-8x faster, which would make multi-step agent loops dramatically cheaper in wall-clock time.

What you could build with it

A startup building browser or computer-use agents could swap its per-step model calls to Mercury 3 and measure task completion time and cost against its current speed-tier model, since each agent step's latency compounds across a multi-step task.

Does it hold up?

Usable now via API (Inception, OpenRouter, Baseten, Vercel) with an OpenAI-compatible interface, but real-world speed is well below the headline figure per OpenRouter's launch-week measurements, so teams should benchmark their own agent loops before switching.

Built with Mercury 3

Inception Mercury 3 explained: speed, benchmarks, price, accesseesel.ai · Independent explainer noting OpenRouter measured a median 160-254 tokens/s in launch week versus Inception's 1,068 claim, and HN testers flagged a slow first token.
Mercury 3 Is 4-8x Faster but Loses Where Agents Need Itagentfred.ai · Critical read: leads 8 of 12 rows in Inception's own benchmark table but loses on agentic coding and long-context recall, and list output price is higher than Claude Haiku 5.5 and GPT-6 Luna.
Inception Releases Mercury 3, Its Flagship Diffusion LLMunite.ai · Launch coverage of the 1M-token context, 40x scale-up over Mercury 1, and launch pricing with 90% cached-input discount.

Learn more

technical deep dive →

First spotted on article: source.

More AI models

OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72

Get the week's best AI launches, plus 3 ideas worth building

One email every Saturday. Ranked by traction, not hype. Free.