Martian Review (2026)

Invented the category, then moved on

Martian is credited with building the first LLM router - forwarding each request to the most suitable model based on performance, cost and reliability, behind an OpenAI-compatible endpoint with cost controls, automatic failover and a free tier of 2,500 requests.

Today, its own site sells something else.

We checked directly, and the homepage describes commercialising interpretability research - “not through consulting or one-off projects, but through products that scale with global LLM usage.” It highlights Thesean AI, described as a lab building best execution for LLMs, and lists research areas including measurement tools such as ARES and Code Review Bench, and theoretical work including K-Steering.

There is no mention of a model router or gateway product.

Why the status is confusing

This is not a clean shutdown, and the ambiguity is the practical problem.

• No shutdown announcement. None that we could find.
• No migration guidance. None published.
• A legacy router landing page appears to still be reachable, which is why comparison sites continue to list Martian among current gateways.

Our reading is that the router has been deprioritised rather than formally discontinued. We are not claiming a shutdown, because nobody announced one.

But for a buyer the effect is the same: there is no current product documentation or pricing to evaluate, so it cannot be shortlisted.

This is the third variation of the same trap this site has documented. Pezzo looked alive because the repository still loaded. Baserun was acquired with the only public signal being a parenthetical in an X bio. Martian looks alive because a landing page nobody removed still resolves.

The check that works is the same in all three cases: look at what the company’s own homepage actually sells. Not a directory listing, not a landing page, not a comparison table - including this one.

If you depend on it

Contact them and get the support position in writing. A URL that resolves is not a commitment.

Then plan the migration assuming the answer may be unsatisfying. The one genuinely good thing here is that the historic router exposed an OpenAI-compatible endpoint, so moving to LiteLLM or OpenRouter is largely a base URL and key change rather than a rewrite.

Do that work before you need it.

The company is fine, and the idea was right

Two things worth separating out, because neither is a failure story.

The company is active and well capitalised - founded in San Francisco around 2022 with roughly $32M raised, now doing substantive interpretability work. This is a pivot, not a collapse.

The routing idea was correct and won. Per-request routing to the most suitable model is now widely accepted - LiteLLM, Portkey, OpenRouter and the cloud vendors’ gateways all do it. What happened is that routing became a feature of gateways rather than a product in itself, which is a common trajectory for a good idea in infrastructure. Martian moving toward interpretability reads as following a more defensible position rather than conceding that routing does not work.

Martian also published RouterBench, a benchmark for multi-LLM routing systems, which remains worth reading if you are evaluating routing approaches generally.

One reported figure we are not repeating

A report from April 2026 claimed Martian was nearing a $1.3 billion valuation. That came from a Medium post rather than a primary source or filing, and we have excluded it from our facts.

Same caution we applied to a HiddenLayer funding figure that proved to be a re-syndicated older round. For anything financial in this category, prefer company announcements or filings - we have already documented two AI security deals where filed consideration was materially below the press-reported number.

Should you use it?

No, not as a gateway. There is no current product to evaluate.

Read RouterBench if routing strategy is something you are actively thinking about.

Bottom line: a pioneering company that moved on from the thing it pioneered, leaving no announcement and a landing page that makes it look otherwise. Use LiteLLM or OpenRouter, and treat this page as a reminder to check what a vendor’s homepage actually sells before shortlisting them.

---

Current product positioning verified directly against the company’s own site on 3 August 2026. No shutdown or discontinuation announcement for the router could be found, and none is asserted here; a legacy landing page may still be reachable. A reported $1.3B valuation from a non-primary source is noted but not relied upon. This is a researched directory entry - we have not instrumented this product with our reference application.