vs
Dolphin vs Boundless
Dolphin sells uncensored models and very cheap DeepSeek V4.1 Flash on consumer GPUs. Boundless offers abliterated models only to vetted security teams.
By The Subconscious Team · Updated
Dolphin vs Boundless: key differences
Dolphin publishes uncensored fine-tunes and serves models on idle consumer GPUs across a decentralized network, with validators checking that nodes run the right model. Its API launched in October 2026 with DeepSeek V4.1 Flash at $0.01 in and $0.07 out per million tokens, publishes no speed, context or SLA, and signs users up through a crypto wallet and Telegram. Boundless charges $0.20 in and $1.00 out for DeepSeek V4.1 Flash with 1M context, which Vercel AI Gateway measures at 121 tokens per second. Its Defensive Access program offers an abliterated DeepSeek V4.1 Flash 1017 by application to vetted, authorized security testers, at $0.24 in and $0.96 out with a listed 500 tokens per second.
Pick Dolphin for the lowest price on latency-tolerant work, or for uncensored models in general products. Pick Boundless when procurement needs standard billing and zero data retention in the US or Europe, or when a security team needs an abliterated model under an authorized program.
What Dolphin and Boundless do
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 profileBoundless
Boundless launched in September 2026 as an inference partner for AI-native startups. It sells two things. The Model API serves pre-tuned models on shared capacity, billed per token from prepaid credits with a budget per API key. Tailored inference reserves capacity for one customer, and Boundless engineers pick the model on the customer’s own evals, tune precision, batching, caching, speculative decoding and parallelism, then run the deployment against uptime, throughput and latency targets written into the contract. A free two-week evaluation tests those targets on real traffic against the customer’s current provider first.
Example models: GLM-5.3, DeepSeek V4.1 Flash, Kimi K3
Full Boundless profileShould you choose Dolphin or Boundless?
Dolphin vs Boundless at a glance
| Attribute |  Dolphin | Boundless |
|---|---|---|
| Model access | Open weights, uncensored fine-tunes | Open weights, plus closed GPT and Claude |
| Flagship models | DeepSeek V4.1 Flash, Dolphin Mistral 24B | GLM-5.3, DeepSeek V4.1 Flash, Kimi K3 |
| Speed | Unknown | 342 tok/s on DeepSeek V4.1 Flash 1017 (Vercel) |
| Price | $0.01 in, $0.07 out (DeepSeek V4.1 Flash) | $0.04–$2.30 in, $0.21–$11.40 out per 1M (open) |
| Customization | Own uncensored fine-tunes | Serving-stack tuning, no fine-tuning |
| Deployment | Decentralized serverless API | Serverless API, reserved tailored capacity |
| Long context | Unknown | 1M context on most models |
Frequently asked questions
What is the difference between Dolphin and Boundless?
Dolphin sells uncensored models and very cheap DeepSeek V4.1 Flash on consumer GPUs. Boundless offers abliterated models only to vetted security teams.
When should I choose Dolphin over Boundless?
DeepSeek V4.1 Flash at $0.01 in; Uncensored open fine-tunes; Batch work where latency matters little.
When should I choose Boundless over Dolphin?
Abliterated models for vetted security teams; Measured speed on Vercel AI Gateway; Zero data retention in the US or Europe.
Is Dolphin or Boundless cheaper?
Dolphin: $0.01 in, $0.07 out (DeepSeek V4.1 Flash). Boundless: $0.04–$2.30 in, $0.21–$11.40 out per 1M (open). 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.