Repository navigation
Proposal — keep one always-available, non-frontier free model (no hard 24h cutoffs) #63368
Description
The current state
The free tier today consists of ~13 models, all marked "free for a limited time." Two were deprecated this week alone (Pulse Alpha Free on Nov 6, Crouton Alpha Free on Nov 9). Only LongCat 2.5 Free publishes a concrete limit (100 req/day); the rest have undocumented daily cutoffs that hit without warning or explanation. When a model runs out, there is no fallback — your session just stops.
Who actually uses the free tier
It's easy to frame free-tier users as freeloaders, but that misses the biggest group: people building open source software for the community. They're not hoarding free tokens for personal gain — the output of their usage is public code anyone can use. And crucially, most of them were never going to buy a subscription in the first place: they don't need one for themselves. Cutting the free tier doesn't convert them into paying customers; it just loses them.
But the calculus changes over time. Open source developers become consultants, freelancers, team leads, founders. The day they do need to pay — for work or for their team — they will go where they've been most comfortable. Every hour of good experience on the free tier is an investment in that future decision. The free tier is not charity; it's the front door for an entire segment of users who currently belong to no one. Whoever treats them best while they're broke owns them when they're not.
Why the current design hurts
-
The free tier is the hook. For most people choosing between coding agents, the free tier is the product experience. OpenCode grew on "you can just try it, for real." Right now that means a rotating cast of stealth models until they hit an undocumented wall or get deprecated.
-
Some of these models are already paid for in data. Big Pickle, Iceman Free, Inkling 2 Small, Flint 40B, MiMo-V2.6-Flash and the Nemotron NVIDIA-trial endpoints explicitly retain user data during the free period, and some use it for training. If the provider is already extracting value from usage, the marginal-cost argument for hard daily cutoffs is much weaker. These models should have the widest leash, not frontier-level treatment.
-
Every competitor is more generous.
- Anthropic's free tier uses a 5-hour rolling window. Even at the free level, your quota comes back within the same workday.
- Freebuff serves its lower-quality models (Beam 1.0, Nemotron 3.5 Lightning, Glyph Cluster, Solar Pro 4/Mini 4) simply marked "Free" — no request counter, no visible daily cap. The only metered tier budgets 12 hours of use per day, and even exhausted, the free Standard models remain available as fallback.
- Freebuff's other free models (DeepSeek, Muse Spark) carry generous daily caps, not walls.
- NVIDIA and Alibaba reportedly offer unlimited models on their free endpoints (I haven't verified this personally, but the direction of the market is clear).
Against this landscape, OpenCode's free tier — undocumented limits, 24-hour full cutoffs, models rotated out with days of notice — is the tightest of the group. The gap is not small, and it's the first thing every new user evaluates.
The proposal
Designate one non-frontier model as the permanent free-tier backbone:
- Always available (not "limited time"), with removals announced in advance in the changelog.
- Soft limits instead of hard cutoffs: a rolling window (e.g. 5h, like Anthropic) or a generous time-based daily budget with graceful degradation (slower responses / smaller context) instead of a wall.
- Documented limits, published per model like LongCat's 100 req/day is today.
- Keep the rotating experimental/stealth models as they are — they're great as a bonus, not as the only option.
On freeloaders
Yes, they exist, and abuse is real. But policy shouldn't be designed around the worst users: per-request rate limiting, the OpenCode-client requirement, and account-level measures are already available levers. Honest daily drivers getting silently walled at 2pm — including people shipping OSS with those tokens — is a worse trade for the project's spirit than a few people extracting more free tokens than they "deserve."
The free tier determines a lot of OpenCode's popularity relative to other coding agents. I'd rather it be defined by openness than by churn.
Question for the community: would you use or recommend OpenCode more if there was one always-on, non-frontier free model with a rolling window instead of daily cutoffs? And what limit would feel fair to you?
Activity
valledemena commentedon Nov 11, 2026
Spanish version / versión en español (for the Spanish-speaking community — same content):
Estado actual
El free tier hoy son ~13 modelos, todos marcados como "gratis por tiempo limitado". Solo esta semana se descontinuaron dos (Pulse Alpha Free el 6 de noviembre, Crouton Alpha Free el 9). Solo LongCat 2.5 Free publica un límite concreto (100 req/día); el resto tiene cortes diarios sin documentar que llegan sin aviso ni explicación. Cuando un modelo se agota, no hay fallback — tu sesión simplemente se detiene.
Quiénes usamos el free tier de verdad
Es fácil caricaturizar al usuario del free tier como freeloader, pero eso pasa por alto al grupo más grande: la gente que construye software open source para la comunidad. No acumulan tokens gratis para beneficio personal — el resultado de su uso es código público que cualquiera puede usar. Y lo esencial: la mayoría nunca iba a comprar una subscripción, porque no la necesita para ellos mismos. Recortar el free tier no los convierte en clientes de pago; simplemente los pierde.
Pero el cálculo cambia con el tiempo. El dev de open source acaba siendo consultor, freelancer, tech lead, fundador. El día que sí necesite pagar — por un trabajo o para su equipo — irá a donde ha estado más cómodo. Cada hora de buena experiencia en el free tier es una inversión en esa decisión futura. El free tier no es caridad: es la puerta de entrada para todo un segmento de público que hoy está en tierra de nadie. Quien los trate mejor mientras no pueden pagar, los tendrá cuando sí puedan.
Por qué el diseño actual perjudica
El free tier es el gancho. Para la mayoría de la gente que elige entre coding agents, el free tier es la experiencia del producto. OpenCode creció sobre "puedes probarlo de verdad, ya mismo". Ahora eso significa un elenco rotatorio de modelos stealth hasta que chocas con un muro sin documentar o los retiran.
Algunos de estos modelos ya se han cobrado en datos. Big Pickle, Iceman Free, Inkling 2 Small, Flint 40B, MiMo-V2.6-Flash y los endpoints de prueba de Nemotron (NVIDIA) retienen explícitamente datos de usuarios durante el periodo free, y algunos los usan para entrenar. Si el proveedor ya está extrayendo valor del uso, el argumento de coste marginal para los cortes duros diarios es mucho más débil. Estos modelos deberían tener la correa más larga, no un trato de nivel frontier.
Todos los competidores son más generosos.
Frente a ese panorama, el free tier de OpenCode — límites sin documentar, corte completo de 24h, modelos retirados con días de aviso — es el más estrecho del grupo. La diferencia no es pequeña, y es lo primero que evalúa cualquier usuario nuevo.
La propuesta
Designar un modelo no-frontier como backbone permanente del free tier:
Sobre los freeloaders
Sí, existen, y el abuso es real. Pero la política no debería diseñarse alrededor del peor usuario: el rate limiting por request, el requisito de cliente OpenCode y las medidas a nivel de cuenta ya son palancas disponibles. Que los usuarios honestos amanezcan con muros silenciosos a las 2pm — incluidos los que publican OSS con esos tokens — es un peor intercambio para el espíritu del proyecto que que unos pocos saquen más tokens de los "debidos".
El free tier determina mucho la popularidad de OpenCode frente a otros coding agents. Prefiero que lo defina la apertura que la rotación.
Pregunta a la comunidad: ¿usarías o recomendarías OpenCode más si hubiera un modelo free siempre disponible, no-frontier, con ventana móvil en vez de cortes diarios? ¿Y qué límite te parecería justo?
opencodejr commentedon Nov 11, 2026
Thanks for the detailed proposal, @valledemena!
If I understand correctly, you'd like the free tier to include one permanent, non-frontier free model that:
while the rotating experimental and stealth models stay as an extra. The maintainers will consider it.
Related issues, closest first:
neriousy commentedon Nov 11, 2026
free tier issues here please: #59387