vs
Luminal vs Dolphin
Luminal compiles your own models for GPUs you choose. Dolphin serves one model on a decentralized network of consumer GPUs.
By The Subconscious Team · Updated
Luminal vs Dolphin: key differences
Luminal's open-source compiler turns PyTorch or Hugging Face models into native kernels, sold as early-access serverless inference or as an on-prem license with SLAs. Dolphin's API launched in October 2026 with one model, DeepSeek V4.1 Flash, at $0.01 in and $0.07 out per million tokens, run by GPU owners who earn $POD tokens, with validators checking outputs. Dolphin publishes no speed, context or SLA.
Both are young. Luminal fits teams that bring their own model and care about throughput on controlled hardware. Dolphin fits cheap background calls and uncensored models, where variable latency on consumer GPUs is acceptable.
What Luminal and Dolphin 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
Dolphin
Dolphin began in 2023 as Eric Hartford's series of open-source, uncensored fine-tunes, trained to drop the refusals and baked-in guidelines of their base models so that the developer sets the rules through the system prompt. Current releases include Dolphin Llama 3.1 8B and Dolphin Mistral 24B Venice Edition, built with Venice, and the project publishes nearly 100 models on Hugging Face. It also runs a chat app, a Telegram bot and a synthetic data service that generates answers to commonly refused prompts.
Example models: DeepSeek V4.1 Flash, Dolphin Mistral 24B
Full Dolphin profileShould you choose Luminal or Dolphin?
Luminal vs Dolphin at a glance
| Attribute | Luminal |  Dolphin |
|---|---|---|
| Model access | Small hosted catalog plus bring your own weights | Open weights, uncensored fine-tunes |
| Flagship models | DeepSeek V4.1 Flash, Ideogram 4.5 | DeepSeek V4.1 Flash, Dolphin Mistral 24B |
| Speed | 36K tok/s on GPT-OSS 120B, 8xH100 (vendor) | Unknown |
| Price | $0.1425 in, $0.57 out (DeepSeek V4.1 Flash) | $0.01 in, $0.07 out (DeepSeek V4.1 Flash) |
| Customization | Compiles any PyTorch or HF model | Own uncensored fine-tunes |
| Deployment | Serverless (early access), on-prem license | Decentralized serverless API |
| Long context | Unknown | Unknown |
Frequently asked questions
What is the difference between Luminal and Dolphin?
Luminal compiles your own models for GPUs you choose. Dolphin serves one model on a decentralized network of consumer GPUs.
When should I choose Luminal over Dolphin?
Serving custom or fine-tuned models; Throughput on owned GPUs; On-prem deployments with SLAs.
When should I choose Dolphin over Luminal?
Very cheap DeepSeek V4.1 Flash calls; Uncensored models; Requests nodes cannot log.
Is Luminal or Dolphin cheaper?
Luminal: $0.1425 in, $0.57 out (DeepSeek V4.1 Flash). Dolphin: $0.01 in, $0.07 out (DeepSeek V4.1 Flash). 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.