Inception 发布扩散语言模型 Mercury 3，速度达 1068 tokens/s

Add as a preferredsource on Google

Inception has released Mercury 3, the newest version of its diffusion language model. Mercury 3 is a diffusion LLM from Inception that writes answers in parallel, not token by token. Inception reports 1,068 tokens per second on NVIDIA GPUs and a 1M-token context window. On Inception’s own benchmark table, it leads GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 tests. For developers building agents, voice bots and search tools, this is a speed-first model with competitive scores.

TL;DR

Size: Parameters not disclosed. Inception says it is more than 40x larger than Mercury 1. Context is 1M tokens.

Runs on: Inception-served NVIDIA GPUs (exact GPU not disclosed). API only, no open weights or quantized builds.

Performance: Leads 8 of 12 benchmarks vs GPT-6 Luna and Haiku 5.5 (self-reported, medium settings).

Best: SWE-Bench Atlas 51.08, about 18 points ahead of both rivals.

Worst: DeepSWE v1 46.31, 5.31 points behind Haiku 5.5 (51.62).

What is Mercury 3?

Mercury 3 is Inception’s fourth Mercury release, after Mercury 1, Mercury 2 and Mercury 2.5. Mercury 1 launched in 2025. Mercury 2.5 added tunable reasoning and a 260K context window, per Inception’s launch announcement.

1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec. pic.twitter.com/TYevDl43Lm— Stefano Ermon (@StefanoErmon) October 19, 2026

How does a diffusion LLM work?

Autoregressive models write one token after another. Each new token waits for the previous one. A diffusion LLM starts from a rough draft of the full answer. It then refines every position over a few passes in parallel.

Fewer sequential steps means faster output on the same GPUs. The open question was whether this approach holds up as models grow larger.

Does diffusion scale?

Here Inception’s answer is yes. Mercury 3 has more than 40x the total parameters of Mercury 1 (mini). Average quality across standard benchmarks rose about 3x between those 2 models. Speed stayed above 1,000 tokens per second.

Inception team plots this as a quality vs size curve across 5 Mercury generations.

How does Mercury 3 perform on benchmarks?

All scores below come from Inception’s table. Every model ran at medium settings.

Mercury 3 posts its largest gains on instruction following and agentic coding. IFBench reaches 76.84, 15 points above GPT-6 Luna. SWE-Bench Atlas reaches 51.08, against 33.33 for Luna and 32.26 for Haiku 5.5. GPQA Diamond lands at 89.23.

The losses matter too. Haiku 5.5 wins DeepSWE v1 (51.62 vs 46.31) and MRCR v2 (78.40 vs 75.50). GPT-6 Luna edges ahead on AA-Omniscience-Public (46.94 vs 46.44) and AA-LCR v1.1 (78.30 vs 77.67). Long-context retrieval and longer coding tasks remain the weak spots.

Mercury 3 vs GPT-6 Luna vs Claude Haiku 5.5

FeatureMercury 3GPT-6 LunaClaude Haiku 5.5

DeveloperInceptionOpenAIAnthropic

Generation methodDiffusion (parallel refinement)Not disclosedNot disclosed

ParametersNot disclosedNot disclosedNot disclosed

Context window1M1.05M1M

Max outputNot disclosed128K128K

Input modalitiesNot disclosedText, imageText, image

Output speed (Inception’s measure)1,068 tok/s~130 tok/s~240 tok/s

Price, $/1M in / out$0.20 / $0.70 list ($0.10 / $0.35 launch)$0.10 / $0.50 (short context)$0.10 / $0.50 (prompts up to 100K)

GPQA Diamond89.2386.3684.51

IFBench76.8461.8254.65

SWE-Bench Atlas51.0833.3332.26

DeepSWE v146.3137.7651.62

MRCR v275.5062.7278.40

Speed and benchmark rows: Inception’s table, medium settings. Specs and prices: each vendor’s own docs.

What do early users report?

Inception quotes 4 early-access partners. These are customer statements, not independent evals.

TinyFish ran 50 live Mind2Web web-navigation tasks. Mercury 3 completed 50% of them. Inception says that beat GPT-6.1-Sol but gives no Sol number.

Octen AI reports 0.9 seconds median generation and 2.2 seconds end to end for smart-glasses search.

RevisionDojo says hints and marked answers return in about a second.

Lorikeet says Mercury 3 beat other models on latency and evals for support agents.

How can developers use Mercury 3?

Mercury 3 is live on the Inception API Platform, OpenRouter, Baseten and Vercel. The API is OpenAI compatible. The docs use the model string mercury-3 and a reasoning_effort parameter.

Inception also ships 2 siblings:

Mercury Voice targets time to first answer token in voice agents.

Mercury Decide is a classification model.

Key Takeaways

Mercury 3 is a diffusion LLM claiming 1,068 tok/s and 1M context.

It leads 8 of 12 vendor-run benchmarks vs GPT-6 Luna and Haiku 5.5.

Biggest wins: SWE-Bench Atlas 51.08 and IFBench 76.84.

Haiku 5.5 still wins DeepSWE v1 and MRCR v2.

Launch price is $0.10 / $0.35, rising to $0.20 / $0.70.

Jean-marc Mommessin+ postsBio ⮌

Jean-marc is a successful AI business executive .He leads and accelerates growth for AI powered solutions and started a computer vision company in 2006. He is a recognized speaker at AI conferences and has an MBA from Stanford.

Jean-marc MommessinMeta, OpenAI and Uber Just Taught AI Agents to Talk First. What About When to Stay Quiet?

RELATED ARTICLESMORE FROM AUTHOR

Exa 3.0 Search API Tops Perplexity, Brave and Parallel on 8 Benchmarks

Meet Drex Max: Nace AI’s Multimodal Decision Model for Computer-Use Agents with a Score 67.59

MiniMax Just Launched M3.1 Flash: 428B MoE Agentic Model at $0.15/M Input

NVIDIA Open Sources Run:ai Under Apache 2.0 for Kubernetes GPU Orchestration