Mercury 3
A diffusion-model-based generative AI model focused on scalable deployment. Through engineering optimizations such as few-step sampling, it reduces inference costs and achieves a commercially acceptable balance between throughput and latency.
Timeline (last 90 days)
Inception Labs released Mercury 3, a diffusion LLM, claiming it surpasses GPT-6 Luna and Claude Haiku 5.5 on benchmarks and runs 4 to 8 times faster; generation speed reaches 1,068 tokens/second on NVIDIA GPUs, and the context window expands from 260K to 1 million tokens.
New architectures typically face adaptation issues with toolchains, fine-tuning methods, and deployment frameworks, and information on whether Mercury 3 is compatible with mainstream inference frameworks and fine-tuning paradigms is currently limited.
Inception released Mercury 3, a large language model based on a diffusion architecture.
Official data shows that Mercury 3 reaches an inference speed of 1,068 tokens per second.
According to Inception's own published comparison table, Mercury 3 leads GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 benchmarks.
Mercury 3 offers a context window of 1 million tokens.
To use Mercury 3 in coding agents such as Claude Code, Codex, and Cursor, you need to install the latest version of the Vercel CLI and run npx vercel ai-gateway setup, then select inception/mercury-3 in the agent.
Inception claims that Mercury 3 achieved "strong results" in benchmarks, but the original text did not disclose the specific comparison models and score details.
When calling Mercury 3 on the AI Gateway, you need to set the environment variable AI_GATEWAY_API_KEY and specify the model name as inception/mercury-3.
Inception's Mercury 3 model is now available on the Vercel AI Gateway.
5 more timeline events
All Facts (15)
Inception Labs released Mercury 3, a diffusion LLM, claiming it surpasses GPT-6 Luna and Claude Haiku 5.5 on benchmarks and runs 4 to 8 times faster; generation speed reaches 1,068 tokens/second on NVIDIA GPUs, and the context window expands from 260K to 1 million tokens.
50%UnverifiedAccording to Inception's own published comparison table, Mercury 3 leads GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 benchmarks.
50%UnverifiedNew architectures typically face adaptation issues with toolchains, fine-tuning methods, and deployment frameworks, and information on whether Mercury 3 is compatible with mainstream inference frameworks and fine-tuning paradigms is currently limited.
50%UnverifiedOfficial data shows that Mercury 3 reaches an inference speed of 1,068 tokens per second.
50%UnverifiedMercury 3 offers a context window of 1 million tokens.
50%UnverifiedTo use Mercury 3 in coding agents such as Claude Code, Codex, and Cursor, you need to install the latest version of the Vercel CLI and run npx vercel ai-gateway setup, then select inception/mercury-3 in the agent.
50%UnverifiedInception claims that Mercury 3 achieved "strong results" in benchmarks, but the original text did not disclose the specific comparison models and score details.
50%UnverifiedWhen calling Mercury 3 on the AI Gateway, you need to set the environment variable AI_GATEWAY_API_KEY and specify the model name as inception/mercury-3.
50%UnverifiedMercury 3 comes with a 50% launch-period discount on the AI Gateway.
50%UnverifiedMercury 3 uses diffusion for parallel refinement to generate text, rather than traditional token-by-token autoregressive output.
50%UnverifiedAccording to figures published by Inception, Mercury 3 generates at over 1000 tokens per second.
50%UnverifiedMercury 3 has 75 points and 28 comments on Hacker News.
50%UnverifiedThe specific technical details of Mercury 3 await further official disclosure.
50%UnverifiedInception released Mercury 3, a large language model based on a diffusion architecture.
50%UnverifiedInception's Mercury 3 model is now available on the Vercel AI Gateway.
50%