Mozilla AI opens early access to AI coding subscriptions starting at $10
Otari combines model access, provider routing and tools in one plan. Usage beyond the allowance keeps running against an organization’s balance.
Loading page…
Otari combines model access, provider routing and tools in one plan. Usage beyond the allowance keeps running against an organization’s balance.
Listen to this story
Mozilla AI announced Otari Subscriptions on Nov. 6, 2026, offering limited early access to a single key for open-weight coding models through existing tools. Go costs $10 monthly for $30 of usage; Pro is $40 for $120. Otari routes each requested model to a provider based on price, service health and latency, then keeps a session on that provider for prompt caching. The allowance is not a spending cap: overages draw from an organization balance, while Claude and GPT use is billed separately.
Otari works with Pi, Hermes, Claude Code, Codex and OpenCode, which developers can point to the service instead of configuring separately for each provider.
Every plan includes web search, web fetch and code execution, plus routing settings, budgets and guardrails, all subject to usage allowances.
Mozilla AI says the Otari gateway is open source and can be self-hosted.
Mozilla AI is bundling access to open-weight coding models into a monthly subscription, rather than leaving developers to assemble separate provider accounts and tools. Announced November 6, 2026, Otari Subscriptions starts at $10 a month in limited early access, with one key for supported coding tools and usage allowances rather than unlimited access.
Mozilla AI’s starting argument is that open-weight models can now handle the work coding agents need: planning changes, editing multiple files, running tools and recovering from errors. The company presents that as a reason to reconsider closed-model defaults, not as a benchmark result accompanying this subscription announcement.
Its diagnosis is that capable models alone do not make a convenient service. Developers still have to choose a model and a provider, manage accounts, keys and bills, and configure their coding tools. Mozilla AI also points to tools that must be added separately and per-token charges that make monthly spending harder to predict.
Otari’s response is a single-key subscription for open-weight models, with new models added as they are released. Mozilla AI lists Pi, Hermes, Claude Code, Codex and OpenCode as supported coding tools. Developers point their existing coding tool at Otari rather than configuring it separately for each provider.
Behind that connection, Otari chooses a provider serving the requested model using price, service health and latency—the time a response takes. That routing concerns where a model runs. Mozilla AI describes the service as finding a suitable provider for the model, rather than choosing a different model for each coding task.
Once a session starts, Otari keeps it on the same provider so prompt caching continues to work, according to Mozilla AI. The design combines provider selection with continuity within a session, rather than moving every request wherever the routing criteria might otherwise point.
The Go plan includes $30 of usage for a $10 monthly subscription at early-access pricing.
The Pro plan includes $120 of usage for a $40 monthly subscription at early-access pricing.
Both subscriptions are personal plans covering open-weight models. Compatibility with Claude Code or Codex should not be confused with including the closed models associated with those tools. Mozilla AI says use of closed models such as Claude and GPT is paid separately from the organization’s balance, outside the subscription allowance.
Mozilla AI says it charges no service fee on subscription usage. Exhausting the plan does not stop work: further usage draws from the organization’s balance. The monthly subscription therefore sets the price for an included allowance, not a ceiling on total spending. Continued access and a fixed overall bill are different promises.
The underlying Otari gateway is open source and can be self-hosted, Mozilla AI says, offering a deployment option beyond its hosted service. For the subscription itself, access remains limited: invitations will go out in waves over the coming weeks. A Teams plan is also planned, but the announcement gives no launch date or pricing for it.
Loading discussion...
Join the conversation
Explain what would matter more than convenience.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.