LunaRoute vs Featherless

What “flat price” covers

This is the difference that matters most if you run agents.

Featherless's Chat plan is $25 a month with no per-token billing, but it's for a person chatting. Its terms say individual plans are “for interactive use or proto-typing and experimentation by the purchaser”, and other use can get the subscription terminated without a refund. Its own guide to coding agents puts it plainly: “The Chat plan is for human-driven use only and excludes automation, so don't try to run an agent fleet on it.” (Featherless blog, August 2026)

For Claude Code, opencode or other agents using the shared API, Featherless points you to the Developer plan. That plan starts at $50 a month in credits, and usage is billed per token. GLM 5.3 costs $1.40 per million input tokens and $4.40 per million output tokens, or $0.26 per million for cached input. Featherless only bills successful requests, which is a nice touch, and unused credits carry over without expiring. See Developer and Credits.

The plan limits differ too: Chat has 32K context and four concurrent units; Developer offers up to 256K and 100 units. Those are not 100 simultaneous requests. Each model consumes a number of units while its request runs; the documented examples use one, two or four units per request, with exceptions. Four units can mean four small-model requests or one very-large-model request. Requests above the budget receive HTTP 429. See concurrency limits.

That is the Chat-versus-Developer API comparison, not a claim that every Featherless agent workload must be token-billed. Business offers dedicated hardware at a contracted price. Its Agent Marketplace also describes hosted sandboxes above the entry Chat tier with inference bundled into the subscription. The pricing page includes one agent environment on Developer, while the Developer billing docs explicitly meter API requests. Check the terms for the particular hosted-agent offering; “bundled” should not be read as unlimited Developer API credits.

LunaRoute's coding plans are flat for agents too. You pick how many priority concurrent requests you want, from 2 on Pro ($99) to 10 on Teams ($499.99), and the price for included inference stays the same however much your agents run, within model and plan limits. Requests above your priority allowance can use available overflow capacity automatically; queue limits and timeouts still apply, and requests can be rejected when those limits are reached.

Web search includes 100 searches a month per organization. Above that you opt in to pay or bring your own key, subject to your organization's search limits. External providers and optional add-ons are separate from included inference; see web search.

Which works out cheaper depends on how much your agents run. If they run occasionally, per-token billing can cost less. If they run for hours a day, or you want to stop thinking about the meter, a flat plan is easier to live with. More in Concurrency vs token pricing.

LunaRoute also has Personal Agent at $19.99 a month for one non-priority request at a time, with its own model scope. Featherless's $25 Chat plan is cheaper than our base coding plans, not every LunaRoute plan; compare intended use, models and capacity rather than the entry price alone.

Using it with Claude Code and other coding agents

LunaRoute speaks Anthropic's Messages format natively, so Claude Code connects directly. Setup starts with one command:

npx @lunaroute/cli setup claude-code

The command prints the environment settings to apply; it does not write Claude Code's configuration or your shell profile. Follow the printed instructions, then restart Claude Code. See the setup steps.

The same CLI sets up Codex, opencode, pi, Copilot CLI, Hermes and OpenClaw. See Using Claude Code with LunaRoute.

Featherless offers an OpenAI-compatible API. Agents that speak the OpenAI format can use it directly. For Claude Code, Featherless's own guide routes requests through claude-code-router, a proxy that translates Anthropic-format requests into OpenAI format. That works, but it's one more piece to install and keep running.

Models

Featherless's catalog is much bigger: it advertises over 40,000 open models, including large coding models and many fine-tunes. If you want to try a niche model, or switch models often, that breadth is a real advantage.

LunaRoute serves a smaller, curated set and tunes how each one runs on our own GPUs: GLM 5.3, GLM 5.3 Flash and DeepSeek 4.1 Flash, plus decision, embedding and image models. With open models, the serving setup affects speed and reliability as much as the model name does, so we'd rather run a few models well. Our GLM 5.3 is a vision-enabled NVFP4 build with a 512K context window, and its weights are public on Hugging Face. Featherless describes its GLM 5.3 serving configuration as FP8 with a 256K context on Developer; see its model comparison.

Featherless also documents embeddings and speech generation, alongside its hosted agent environment.

Privacy

Both services say they don't keep API inference prompts or completions. Featherless states that it “does not log chats, prompts, or completions sent through our API” (privacy docs). Its privacy policy retains account, payment and model-usage information and reserves the right to store sampler settings. The API policy is not a blanket promise about hosted-app storage: its Marketplace docs say sandbox files and configuration are preserved.

On LunaRoute, prompts and responses for LunaRoute-hosted inference are processed in memory and discarded by default. We retain account, usage and billing metadata. Generated images and URL-based document artifacts can be retained for up to seven days; third-party traffic, including BYOK, follows the upstream provider's policy. See zero data retention.

Switching from Featherless

Both use API keys and an OpenAI-compatible endpoint, so most clients need a new base URL, key and model ID: use https://gw.lunaroute.com/v1 with your lr_ key and a model from the LunaRoute catalog.

For Anthropic clients that append /v1/messages themselves, use the gateway root, https://gw.lunaroute.com, rather than the OpenAI /v1 base URL. LunaRoute accepts Anthropic Messages at /v1/messages and Responses requests at /v1/responses. For coding agents, the CLI generates the appropriate configuration; Claude Code setup prints settings to apply, and connecting directly to LunaRoute removes the need for claude-code-router.