MindsDB Review 2026: It's MindsHub Now
If you searched for MindsDB expecting the “AI layer for your database” that let you write machine learning in SQL, the honest answer comes first: that is not what ships anymore. mindsdb.com redirects to mindshub.ai, and the product behind the redirect is a different one. Here's what it actually is, what it costs, and whether the thing you came for still exists.
What Changed — Verified September 2026
- • mindsdb.com issues a 301 redirect to mindshub.ai.
- • github.com/mindsdb/mindsdb redirects to github.com/mindsdb/mindshub.
- • The commercial product is now a model router / unified inference platform, not a SQL ML engine.
- • Billing changed from plans to a published pay-as-you-go rate card with no subscription.
- • The SQL query engine survives as a separate open-source documentation surface.
Quick Verdict
Best for: Teams that want one API across many models with a rate card they can actually read, and a real free allowance to build against. The rating is held down by the rebrand, not by the product — as a router MindsHub is solid; as a destination for everyone still searching “MindsDB”, it is a different product wearing the redirect.
What MindsHub Actually Is
MindsHub puts every major model behind one API and one console. You send a request, it routes to the model you named, and if that model is unavailable it falls through a published chain of alternatives so the request still completes. Billing is per million tokens against a rate card that separates input, output, cached input and cache write, and there is no plan to subscribe to.
Three cost paths exist side by side. The free allowance gives you 5M tokens a month on the MindsHub Air alias with no card. Your keys lets you run any model on your own provider accounts, paying the provider directly and using MindsHub purely for routing and console. Credits covers everything else at the published rates.
What is gone from the commercial product is the reason most people knew the name. MindsDB's pitch was machine learning in SQL — connect a live data source, train and query models with the language your analysts already use. That engine still exists as open source on its own docs surface, but it is no longer the product the company sells, and you should evaluate it as a community project rather than as a supported platform.
Built a model router or inference platform? People land on this review while they are still choosing one.
Add it to the data category — a free listing publishes after review, and it is the same page ChatGPT, Perplexity and Google read when a developer asks which LLM router to use. Want it live in minutes with a Verified badge instead? That option is on the form, one-time, no subscription.
MindsHub Pros & Cons
✓ Pros
- •The rate card is fully published and per-model, in dollars per million tokens, with input, output, cached input and cache-write priced separately. In a category where 'contact us' is common, being able to model your bill before signing up is the single most useful thing about it
- •A real free allowance with no card: 5,000,000 tokens a month on the MindsHub Air alias, usable both in the console and through the API. That is enough to build and test something genuine, not a demo quota
- •Published failover chains per model — each entry states which models are tried, in order, if the primary is unavailable. Routers usually treat failover as an internal detail; exposing it means you can reason about what your users will actually get during an outage
- •Bring-your-own-provider-keys is supported: run any model on your own provider accounts and pay the provider directly rather than MindsHub. That removes the usual router objection, which is paying a margin on top of a price you could get yourself
- •No subscription at all — MindsHub Cowork and Unified Inference share the same rate card and there are no plans. For bursty or seasonal workloads that is strictly better than a monthly floor you underuse
- •Long-prompt pricing is handled honestly: models that surcharge beyond a context threshold say so with the exact rates, and models that don't say 'no surcharge — the full context window bills at the standard rates'. Both statements are on the same page
- •Reasoning effort is exposed as a per-task dial on the models that support it, with the vendor stating its own default, so you can trade cost against depth explicitly rather than discovering the default in a bill
- •The original open-source SQL query engine survives as a separate surface, so the thing people originally adopted MindsDB for has not been deleted — just decoupled from the commercial product
✗ Cons
- •The rebrand is the headline risk, not a footnote. mindsdb.com 301s to mindshub.ai and the GitHub org redirects too, which means every tutorial, Stack Overflow answer, course module and internal runbook referencing MindsDB now points at a different product with a different purpose
- •The product category changed, not just the name. 'AI layer for your database, ML in SQL' and 'unified model router' are not the same thing. If you came for the former, the rebrand is effectively a discontinuation with a redirect attached
- •Pay-as-you-go with no subscription cuts both ways: there is no spend ceiling built into a plan, so cost control is entirely on you. An agent loop that misbehaves bills at the rate card until you notice
- •Router margin is real on the credits path. You pay MindsHub's published rate, which sits above what you would pay the provider directly — the bring-your-own-keys option exists precisely because of this, and using it gives up part of the reason to route at all
- •The free MindsHub Air allowance is tied to an alias whose underlying model the vendor explicitly says can move, along with its rates. That is honest disclosure, but it means the thing you benchmarked is not guaranteed to be the thing you ship on
- •Failover means a request can be served by a different model than you asked for. For chat that is a graceful degradation; for structured output, tool calling or anything with model-specific prompt tuning, a silent substitution is a correctness risk you have to design around
- •Every router adds a hop, and the added latency and the extra party in your data path are both real. For a single-model workload with a direct provider contract, the router is overhead with no offsetting benefit
- •The SQL engine's move to a separate open-source docs surface leaves its roadmap and support status ambiguous in a way the commercial product's is not — which is a hard position to build a production data pipeline on
MindsHub Pricing 2026
Read off mindshub.ai/pricing in September 2026. The vendor's own framing is “free to use, pay for what you run” — MindsHub Cowork and Unified Inference share one rate card, and there is no subscription or plan of any kind.
Free
- •5,000,000 MindsHub Air tokens / month
- •No card required
- •Works in the console and the API
- •Rate card applies beyond the allowance
Building and testing before any spend commitment
Your keys
- •Any model on your own provider accounts
- •You pay the provider, not MindsHub
- •Routing and console on top
Teams with existing provider contracts or committed spend
Credits
- •Published per-model rate card
- •Separate input / output / cached / cache-write rates
- •Long-prompt rates stated per model
- •No subscription, no plan
Everything else — bursty, multi-model or experimental workloads
Sample Of The Published Rate Card
| Model | Provider | Input / 1M | Output / 1M | Note |
|---|---|---|---|---|
| MindsHub Air | MindsHub | $0.20 | $1.20 | First 5M tokens a month free |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | No long-prompt surcharge |
| Claude Sonnet 5 | Anthropic | $2.00 | $10.00 | No long-prompt surcharge |
| Claude Opus 5 | Anthropic | $5.00 | $25.00 | No long-prompt surcharge |
| GPT-6 Astra | OpenAI | $10.00 | $50.00 | Surcharge past 272K tokens |
A sample only — the live card covers far more models and also prices cached input and cache write per model. Figures verified September 2026; model prices move, so confirm against the live card before committing a budget.
If You Came For The SQL Engine
MindsDB earned its audience on a genuinely distinctive idea: connect your live data sources, then train, query and serve models using SQL, so the people who already understood the data could do the machine learning without a separate stack. That is why it is in so many tutorials and internal runbooks, and it is why the redirect is jarring.
The engine has not been deleted — it continues as an open-source surface with its own documentation, separate from the commercial product. But separate is the operative word. Its roadmap, release cadence and support status are no longer tied to the company's revenue, which is a materially different risk profile from the one you signed up for if you built a production pipeline on it. Before extending anything that depends on it, check the repository's recent commit and release activity yourself.
And if you are doing an inventory: anything in your codebase, CI, docs or bookmarks pointing at mindsdb.com or the old GitHub path is now a redirect to a different product. Redirects are not permanent contracts. Pin the URLs you actually depend on.
Frequently Asked Questions
Is MindsDB still available in 2026?
Not under that name, and not as the same product. Checked in September 2026: mindsdb.com issues a 301 redirect to mindshub.ai, and github.com/mindsdb/mindsdb redirects to github.com/mindsdb/mindshub. The company rebranded to MindsHub and the commercial product is now a model router and unified inference platform — every major model behind one API and one console, billed from a published per-million-token rate card. The original SQL query engine, the thing most people meant when they said MindsDB, continues as a separate open-source documentation surface rather than as the headline product.
How much does MindsHub cost?
There is no subscription and there are no plans — the vendor states this directly. Three paths: a free allowance of 5,000,000 MindsHub Air tokens per month with no card required, usable in the console and through the API; bring-your-own-keys, where you run models on your own provider accounts and pay the provider rather than MindsHub; and credits, where everything else bills at the published rate card. Rates are per million tokens with input, output, cached input and cache write priced separately, and long-prompt surcharges are stated per model. MindsHub Cowork and Unified Inference share the same rate card.
What happened to MindsDB's 'machine learning in SQL' feature?
It was the product's defining idea — connect data sources live and build predictions and semantic search pipelines entirely in SQL — and it is no longer what the company sells. The engine survives as a separate open-source surface with its own documentation, so the code did not vanish, but it is now decoupled from the commercial roadmap. If SQL-native ML is what you need, treat the surviving engine as an open-source project to evaluate on its own merits rather than as a supported commercial product, and check its recent commit activity before you build on it.
What is MindsHub Air and should I build on the free tier?
Air is an alias MindsHub keeps pointed at whichever model currently strikes the best balance of cost, speed and intelligence — and the vendor is explicit that both the target and its rates can move as better options ship. It is priced at $0.20 per million input and $1.20 per million output tokens, with the first 5M tokens each month free. That makes it an excellent way to build and evaluate without spend. It makes it a weaker choice for a workload with tight output-format requirements, because the model behind the alias is not a fixed target. Prototype on Air, then pin a specific model before anything ships to users.
How does MindsHub compare to OpenRouter?
They occupy the same slot — a single API in front of many providers, with per-model published pricing — so the comparison comes down to details rather than category. MindsHub's distinguishing choices are the stated failover chain per model, a genuinely usable free allowance on its own Air alias, a bring-your-own-keys path, and a rate card that breaks out cached input and cache write alongside input and output. Against that, OpenRouter has the larger model catalogue and far more operational history under its current name, and MindsHub carries the specific risk of a brand that changed identity this year. If you are picking today, price your actual token mix against both rate cards rather than trusting either headline number.
Should I worry about the failover behaviour?
You should design for it rather than worry about it. Publishing the failover chain is a point in MindsHub's favour — most routers do not tell you what happens when your model is unavailable. But the consequence is that a request can be answered by a different model than the one you asked for. For conversational output that is a good trade; for structured JSON, tool calling, or prompts tuned to one model's quirks, a silent substitution can produce output your parser rejects. If that matters, validate the response shape on your side and log which model actually served each request, so a failover shows up in your telemetry rather than in a support ticket.
Who should not use MindsHub?
Three groups. Anyone who came here for MindsDB's SQL-native machine learning should stop and evaluate the surviving open-source engine separately, because the commercial product no longer does that job. Anyone running a single model at steady volume with a direct provider contract is adding a hop, a margin and a party to their data path for no benefit. And anyone who needs a hard spend ceiling should note there is no plan to cap you — pay-as-you-go with no subscription means budget control is entirely your own instrumentation.
Want To Try MindsHub?
5,000,000 free tokens a month on the Air alias, no card required — enough to build something real before spending anything.
Or explore alternatives:
Affiliate disclosure: Some links on this page are affiliate links. If you sign up through them, AISO Tools may earn a commission at no extra cost to you. This never affects our rankings or reviews.
📬 Get the best new AI tools delivered weekly
One concise email with fresh launches, trending picks, and featured standouts.
Join thousands of professionals who discover the best AI tools every week. No spam — unsubscribe anytime.