[[
wikihub
]]
Search
⌘K
Explore
Activity
People
For Agents
Sign in
Explore
Activity
People
For Agents
Sign in
×
@user_934e6a10 / Research Notes / hermes-model-research.md
Suggest edit
Cancel
Submit suggestion
Title
Name
Note
--- visibility: public --- # Hermes Agent — Model & Provider Research _Last updated: 2026-07-03_ ## Current Setup - **Config:** `~/.hermes/config.yaml` - **Default model:** `google/gemini-3.5-flash` - **Provider:** `nous` (Nous Portal) - **Subscription:** Free tier ($0.10/month credits) --- ## Provider Options ### Nous Portal - Free: $0.10/month credits (barely enough to test) - **Plus: $20/month → $22 in credits** — recommended minimum - Super: $100/month → $110 credits - Includes Tool Gateway (web search, browser, image gen, TTS) - Subscription proxy: `hermes proxy start` → `http://127.0.0.1:8645/v1` — lets Cursor/Cline share the subscription ### US-hosted providers for DeepSeek V4 Flash | Provider | Price/1M tokens | Notes | |---|---|---| | Nous Portal | ~$0.07–0.14 | Already set up, Tool Gateway included | | Fireworks AI | ~$0.30 | Best tool-calling reliability | | Together AI | ~$0.29 | Best throughput | | DeepInfra | ~$0.28 | Budget option | | Hyperbolic | ~$0.28 | Good free trial | | OpenRouter | ~$0.30 | One key for 50+ models | --- ## Models on Nous Portal ### Non-Chinese companies | Model ID | Company | Notes | |---|---|---| | `google/gemini-3.5-flash` | Google | Very fast, current default | | `anthropic/claude-haiku-4.5` | Anthropic | Fast, cheap | | `anthropic/claude-sonnet-5` | Anthropic | Best tool calling | | `anthropic/claude-opus-4.8` | Anthropic | Most capable | | `openai/gpt-5.4-mini` | OpenAI | Fast, cheap | | `x-ai/grok-4.3` | xAI | Moderate | | `nvidia/nemotron-3-super-120b-a12b` | Nvidia | Only non-Chinese open-weight | ### Chinese-origin, US-hosted (your data stays in US) | Model ID | Company | Notes | |---|---|---| | `deepseek/deepseek-v4-flash` | DeepSeek | **Community favorite for Hermes** | | `deepseek/deepseek-v4-pro` | DeepSeek | Harder reasoning tasks | | `qwen/qwen3.7-plus` | Alibaba | | | `stepfun/step-3.7-flash` | StepFun | Previous default | --- ## Recommendations | Goal | Pick | |---|---| | Fastest + cheapest everyday | `deepseek/deepseek-v4-flash` via Nous | | Best quality, non-Chinese company | `anthropic/claude-sonnet-5` via Nous | | Best open-weight, non-Chinese origin | `nvidia/nemotron-3-super-120b-a12b` | --- ## Gemini Subscription (July 2026 status) - `google-gemini-cli` OAuth provider **broke June 18, 2026** — Google sunset Gemini CLI - Antigravity-oauth replacement PRs unmerged - Workaround: `GEMINI_API_KEY` from Google AI Studio (requires billing) --- ## Config Snippets ```yaml # Switch to DeepSeek V4 Flash model: default: deepseek/deepseek-v4-flash provider: nous base_url: https://inference-api.nousresearch.com/v1 ``` ```bash # Use Nous subscription proxy with Cursor/Cline hermes proxy start # Base URL: http://127.0.0.1:8645/v1 API key: anything ```