vs
Luminal vs Inception
Luminal compiles any model into faster GPU code. Inception gets speed from a different model design, diffusion LLMs it trains and serves itself.
By The Subconscious Team · Updated
Luminal vs Inception: key differences
Luminal's open-source compiler lowers models to native kernels ahead of time, and it reports GPT-OSS 120B at 36K tokens per second in aggregate on 8 H100s. Luminal Cloud is early access with a tiny priced catalog, and an on-prem license adds SLAs. Inception's Mercury models generate many tokens in parallel on standard GPUs. Mercury 2.5 offers reasoning, tool use and 260K context at $0.20 in and $0.75 out. Inception reports 1,100+ tokens per second; Artificial Analysis measures about 640.
Luminal fits teams that want to serve their own or fine-tuned models faster on hardware they control. Inception fits teams that want a fast hosted model today with public prices, and can accept a closed model from one lab.
What Luminal and Inception do
Luminal
Luminal builds an inference compiler. Where vLLM and SGLang interpret a model at runtime, Luminal compiles it ahead of time into native kernels for GPUs and ASICs. Models get lowered to a small graph of 15 primitive ops, and the compiler searches over fusion, tiling, memory and scheduling choices instead of relying on hand-written rules, which it says can find optimizations like Flash Attention on its own. The compiler is open source in Rust under Apache 2.0 or MIT, runs on CUDA and Metal with ROCm on the roadmap, and works as a torch.compile backend.
Example models: DeepSeek V4.1 Flash, Ideogram 4.5
Full Luminal profile
Inception
Inception builds diffusion large language models (dLLMs). Every other LLM on this list generates text one token at a time; Inception's Mercury models start from a rough draft of a whole block and refine it over a few passes, producing many tokens in parallel. That ties speed less to output length and lets the models run fast on ordinary NVIDIA GPUs rather than custom chips. The company was founded by professors from Stanford, UCLA and Cornell behind core work on diffusion, Flash Attention and DPO, and is led by Stanford's Stefano Ermon. It raised a $50M seed in November 2025 led by Menlo Ventures, with Mayfield, M12, NVentures, Snowflake and Databricks, plus Andrew Ng and Andrej Karpathy as angels.
Example models: Mercury 2.5, Mercury 2
Full Inception profileShould you choose Luminal or Inception?
Luminal vs Inception at a glance
| Attribute | Luminal |  Inception |
|---|---|---|
| Model access | Small hosted catalog plus bring your own weights | Closed weights (Mercury dLLMs) |
| Flagship models | DeepSeek V4.1 Flash, Ideogram 4.5 | Mercury 2.5, Mercury 2 |
| Speed | 36K tok/s on GPT-OSS 120B, 8xH100 (vendor) | 1,100+ tok/s (vendor); ~640 measured |
| Price | $0.1425 in, $0.57 out (DeepSeek V4.1 Flash) | $0.20 in, $0.75 out (Mercury 2.5) |
| Customization | Compiles any PyTorch or HF model | Fine-tuning on request |
| Deployment | Serverless (early access), on-prem license | Serverless API, Bedrock, Azure, private |
| Long context | Unknown | 260K |
Frequently asked questions
What is the difference between Luminal and Inception?
Luminal compiles any model into faster GPU code. Inception gets speed from a different model design, diffusion LLMs it trains and serves itself.
When should I choose Luminal over Inception?
Serving custom or fine-tuned models; High aggregate throughput on owned GPUs; On-prem deployments with SLAs.
When should I choose Inception over Luminal?
Fast hosted output with public prices; Code autocomplete and voice agents; Buying through Bedrock or Azure.
Is Luminal or Inception cheaper?
Luminal: $0.1425 in, $0.57 out (DeepSeek V4.1 Flash). Inception: $0.20 in, $0.75 out (Mercury 2.5). The cheaper choice depends on the model and workload.
Related comparisons
Running long-horizon agents?
If your agents run past 200K tokens, compare both against Subconscious. Our inference stack treats a long-horizon trace as the primary workload, so speed, cost, and accuracy hold up deep into the trace.