The model-provider experience is splitting fast: some people want “one API key for every model and one bill,” some want pure pay-as-you-go with no monthly commitment, and some already bought a subscription somewhere and just want to plug it into Hermes. On August 29, three new providers joined the party on the same day: Ramp Router (router.com’s LLM gateway), Nebius Token Factory (Nebius pay-as-you-go inference), and Tencent TokenPlan (the Tencent Cloud Hunyuan token-plan subscription) — the last one also brings the hy4-preview model into the official catalog. All three PRs (#97915, #97916, #97917) are merged on main.

The three in one line each

Provider	What it is	Who it’s for
Ramp Router	One OpenAI Responses-compatible endpoint; server-side routing to OpenAI/Anthropic/xAI/Fireworks and more, with unified fallbacks and spend controls	Teams that want one key for all models and a single bill
Nebius Token Factory	Nebius pay-as-you-go inference, OpenAI-compatible	Developers with spiky usage who don’t want subscriptions
Tencent TokenPlan	Tencent Cloud Hunyuan token-plan subscription (Anthropic-compatible endpoint), billed by monthly plan	Users who already hold a Tencent Cloud TokenPlan

1. Ramp Router: the gateway provider

The LLM gateway built by Ramp (yes, the fintech company). One endpoint — https://api.router.com/v1 — and one key; requests are routed server-side to OpenAI, Anthropic, xAI, Fireworks, and others, with fallbacks and spend controls handled for you. No more maintaining multiple upstream keys and fallback logic on your side.

Details worth knowing about the Hermes integration:

• The Responses API is the native wire (api_mode="codex_responses" is mandated): Router’s /v1/responses is the real interface; /v1/chat/completions is only a minimal compatibility shim added in August, so Hermes stays on the native line;
• The model catalog is account-scoped: valid model IDs come from your key’s GET /v1/models (BYOK accounts see extra entries), so the official profile ships no hardcoded fallback models — what you see in the picker is fetched live;
• Strict reasoning-effort validation: Router validates reasoning.effort against each model’s declared vocabulary and rejects unsupported combinations with HTTP 400. Hermes caches Router’s catalog capabilities and clamps before sending;
• Requests carry the Hermes-Agent User-Agent.

The plugin (router-provider) is authored by Neel Patel from Ramp, with live verification against api.router.com done in August 2026.

2. Nebius Token Factory: pay-as-you-go inference

Nebius’s “Token Factory” is its pay-as-you-go inference service, OpenAI-compatible. The nebius-token-factory-provider plugin is maintained by Nous Research itself.

One implementation detail worth knowing: some Nebius models expose reasoning-effort support, but not all of them do. Hermes ships a conservative allowlist — only models on it (deepseek-r1/v4, deepseek-reasoner, gpt-oss, glm-5, kimi-k2, minimax-m2, …) can carry reasoning-effort parameters; everything else is clamped automatically, so you never hit errors from passing unsupported parameters.

3. Tencent TokenPlan: subscription + hy4-preview

Tencent Cloud Hunyuan TokenPlan is a monthly token subscription, served over the Anthropic message format (anthropic_messages transport). In Hermes it exists as a built-in overlay provider:

• Internal id: tencent-tokenplan; aliases tokenplan / tencent-lkeap;
• Base URL: https://api.lkeap.cloud.tencent.com/plan/anthropic;
• Environment variable: TOKENPLAN_BASE_URL (the plan’s access endpoint/credentials are injected through it);
• Picker display name: “Tencent TokenPlan”.

The same PR also registers tencent/hy4-preview (Tencent Hunyuan 4 preview) in the official model catalog — TokenPlan subscribers can pick it directly in the model selector.

How to use them

None of these require hand-editing config — hermes setup’s model flow (select_provider_and_model) auto-detects new plugins under plugins/model-providers/ and dispatches to the standard API-key configuration:

hermes setup        # enter model configuration
# choose Ramp Router / Nebius Token Factory / Tencent TokenPlan
# paste the service's API key / access credential when prompted
hermes model        # switch providers anytime afterwards

The built-in Tencent TokenPlan works the same way: set TOKENPLAN_BASE_URL (and the plan credential), pick Tencent TokenPlan in the selector, and choose tencent/hy4-preview as the model.

Caveats

• All three PRs landed August 29 and are currently on main only — not in any release tag (v0.20.6 was tagged August 27); run hermes update first to see these providers;
• Ramp Router’s model list depends entirely on your key’s live catalog — refresh the model list in the picker before first use;
• For subscription plans like TokenPlan, double-check the quota semantics (monthly token allowance vs. metered top-up) on the official Tencent Cloud pages.

Summary

A gateway provider (Ramp Router), a pay-as-you-go provider (Nebius Token Factory), and a subscription provider (Tencent TokenPlan) — three of the most common ways teams buy model access, plus hy4-preview joining the catalog. Every integration goes through the standard hermes setup flow, no hand-written config. To go deeper on model-side configuration, see our model overrides guide and GLM-5.3-Flash guide; the full hermes model reference lives on the command docs page.