Telluvian vs Martian: LLM routing compared
Telluvian and Martian's router, Ship from its lab Thesean, both start from a quality bar and look for a cheaper way to meet it. Ship works with a short list of frontier models: you name one with the ship-like/ prefix and it serves the request at that model's level for less. Telluvian chooses across 190+ models for each prompt, so an easy request can go to a much cheaper model, and it can add per-token hallucination scores.
Last reviewed
At a glance
| Telluvian | Martian | |
|---|---|---|
| Model selection method | Prompt-aware. Model Select predicts each model’s performance on the specific prompt and picks the cheapest one that clears your xPerf quality bar. | You name a base model with the ship-like/ prefix. That model sets the quality bar, and Ship finds the most cost-effective way to meet it, using inference-time optimisation such as steering vectors and prompt optimisation. |
| Models and providers | 190+ models from OpenAI, Anthropic, Google, Qwen, DeepSeek, xAI, Z-AI, Moonshot and Meta, behind one OpenAI-compatible API. | A short list of frontier models, including GPT-5.6 Sol, Claude Opus 4.8, Claude Sonnet 5 and Claude Haiku 4.5, through OpenAI- and Anthropic-compatible APIs. |
| Pricing model | Prepaid, pay as you go. Provider list price for tokens, plus $0.05 per 1M input tokens for Model Select and $1.00 per 1M completion tokens for hallucination scores (optional). | Rates not published. Claims a 50% lower LLM bill without changing model behaviour, guaranteed by an SLA; the quality SLA is arranged with its team. |
| Hallucination detection | Built in. Per-token hallucination scores from probes reading an open-weight proxy model, including for closed models. Off by default. | None documented. |
Key differences
Scope
Ship optimises how a named frontier model is served; Telluvian chooses which of 190+ models answers each prompt.
Quality bar
with Ship it is always the model you name; with Telluvian it is xPerf, a model name or a number on a fixed scale.
Pricing
Thesean does not publish rates and arranges an SLA through its team; Telluvian publishes per-token fees.
Hallucination detection
Telluvian can score every generated token; Ship does not document an equivalent.
Frequently asked questions
What is Thesean?
Thesean is the lab behind Martian's router. Its product, Ship, is an endpoint you call by adding ship-like/ to a supported model name; it then serves the request as cheaply as it can while keeping that model's level of quality.
How do Ship and Telluvian decide what answers a prompt?
Both take a quality bar. With Ship, the bar is always the model you name, and Ship optimises how the request is served to meet it. With Telluvian, the bar is xPerf: a model name, or a number on a fixed scale where the default is 0.9. Model Select then predicts how well each of 190+ candidate models will answer the specific prompt and picks the cheapest one that clears it.
Which models can each use?
Ship's docs list a handful of frontier models, including GPT-5.6 Sol, Claude Opus 4.8, Claude Sonnet 5 and Claude Haiku 4.5. Telluvian's Model Select chooses among 190+ models from OpenAI, Anthropic, Google, Qwen, DeepSeek, xAI, Z-AI, Moonshot and Meta.
Which is cheaper?
Thesean does not publish rates; it claims a 50% lower bill, backed by an SLA you arrange with its team. Telluvian publishes every fee: model tokens at provider list price, plus $0.05 per 1M input tokens when Model Select picks the model. Because Telluvian can send an easy prompt to a far cheaper model, its savings are not capped by the price of the model you named.
Do both use mechanistic interpretability?
Yes, for different jobs. Thesean says Ship's optimisations, such as steering vectors, come from its interpretability research. Telluvian uses probing classifiers on an open-weight proxy model's activations to score every generated token for hallucination.
Try it on your own prompts
Point an OpenAI SDK at Telluvian, send telluvian/gallery-1 as the model, and see which model answers each request. Questions about your use case go straight to the team.
Sources
Competitor details come from their own public docs and pricing pages, checked on . Where a page did not say, neither do we. Spotted something out of date? Tell us at hello@telluvian.ai.