origin not confirmed
Solheim sells 2 different plans, easy to confuse. Two products that look alike and are not. Coder is a flat coding plan: one model (Qwen3.8-27B), unlimited tokens under a fair-use policy aimed at interactive sessions rather than 24/7 agents, at EUR 15 for a 128k window or EUR 30 for 256k. Instances is the actual product the company leads with — you rent capacity rather than consumption, sized on two dials: how many requests run at once (1 to 10) and how much context each gets. Pick Coder if you live in a coding agent and one request at a time is enough; pick an Instance if you need parallelism or a different model, or if you are putting a backend behind it. Read the price carefully, because the two overlap in a way that favours you: Coder gives you Qwen3.8-27B for EUR 15, and renting an instance of the SAME model is EUR 30. The instance price is set per model (EUR 15 / 30 / 45 for Qwen3.6-35B-A3B, Qwen3.8-27B and DeepSeek V4 Flash), not by the instances x context arithmetic the page describes. And the EUR 15 at 64k that the pricing page shows is not an offer: it is the markup a visitor without JavaScript sees, as the page itself admits in an HTML comment. The real self-serve context steps are 128k and 256k.
Jump to: Coder / Coder+ / Coder++ (ninguna) · Instancias (VPL) (concurrencia)
| Tier | $/month | Included usage |
|---|---|---|
| Coder | $15.00 | — |
| Coder+ | $24.00 | — |
| Coder++ | $48.00 | — |
console only units unstated
As advertised. "Unlimited, fair use policy" en los tres. Lo que cambia es la ventana y la concurrencia: Coder 128K y 1 peticion en vuelo; Coder+ 256K y 1 en vuelo; Coder++ 256K y 3 simultaneas. El desbordamiento NO falla: "calls queue instead of failing. No 429s, and a queued request never costs extra". Anual con hasta un 20% menos: 150 / 240 / 480 EUR al ano.
What that actually means. No hay unidad que convertir porque no hay contador: "no hard token cap and no meter". Lo que limita es el CAUDAL, no el volumen — una peticion a la vez. El uso razonable esta dicho con una cifra, cosa rara en este mercado: "if you keep using it for more than 14 hours per day look at our full instances", y declara que el plan apunta a sesiones interactivas, no a agentes 24/7.
Where to check what is left. consola del proyecto
Models covered. qwen3.8-27b
checked 2026-10-02 · high (precios y mecanica, leidos de la pagina y de su config.json) · source 1 source 2 source 3
| Tier | $/month | Included quota |
|---|---|---|
| Starter | $None | — |
| Rise | $None | — |
| Plus | $None | — |
console only quota converts to units
As advertised. SIN TOPE DE USO, y es el argumento del producto: "no request quota", "nothing counts down and nothing resets", "inside that size, run as much as you like, all day". Lo que se alquila es CAPACIDAD, no consumo. Los dos unicos limites son estructurales y los fijas tu: cuantas peticiones a la vez (una "instancia" es una peticion en vuelo, de 1 a 10) y cuanta ventana de contexto cada una. Pasos reales de contexto 128K y 256K, dos pasos en autoservicio. Por encima de 10 instancias o de 128K pasa a tarjeta dedicada, con precio contra el hardware. NO lleva la clausula de uso razonable del plan Coder: esa es solo del plan, y su propia FAQ manda AQUI a quien la pase ("if you keep using it for more than 14 hours per day look at our full instances"). El tope de uno es la puerta de entrada del otro.
What that actually means. Categoria distinta a todo lo demas de esta tabla: no se compra consumo, se compra capacidad. "Rent the machine. Not the tokens." No hay tokens, ni peticiones, ni creditos, ni ventana que se reinicie, asi que no hay nada que convertir: el limite es cuantas peticiones a la vez y cuanto contexto cada una, y las dos cifras las pones tu. Los tokens se cuentan pero no se facturan.
Where to check what is left. consola, por proyecto
Models covered. qwen3.6-35b-a3b, qwen3.8-27b, deepseek-v4-flash-0731
checked 2026-10-02 · high (tarifa leida del config.json del propio sitio) / none (escalado por instancias) · source 1 source 2