Free Model Rankings

This page ranks the hosted models that Hermes Router uses by default on a free tier, free trial endpoint, or included monthly credit. It does not include OpenAI, Anthropic, Codex, Kimi, or OpenCode Go because those require paid API usage or a subscription. Local models are separate because their quality and speed depend on the model and hardware you choose.

Last verified: July 31, 2026.

Quick picks

Need	Start with	Provider
Best overall quality	deepseek-v4-flash-free	OpenCode Zen
Strong agent reasoning	nemotron-3-ultra-free	OpenCode Zen
Multimodal work	mimo-v2.5-free	OpenCode Zen
Fastest reasoning endpoint	openai/gpt-oss-120b	Cerebras
Stable general-purpose model	mistral-medium-3-5	Mistral
Coding on a smaller model	north-mini-code-free	OpenCode Zen
Large multimodal context	gemini-2.5-flash-lite	Gemini

Quality ranking

The order is based primarily on the Artificial Analysis Intelligence Index, which combines reasoning, knowledge, coding, tool-use, and long-context evaluations. The score measures model quality, not endpoint speed or free quota size. A higher score is better. Scores can change when the benchmark suite is updated.

Rank	Model	Hermes providers	AA score	Best for	Free-tier note
1	deepseek-v4-flash	NVIDIA NIM, OpenCode Zen	40	Coding, reasoning, long agent runs	NVIDIA trial endpoint; OpenCode promotion
2	nemotron-3-ultra	OpenCode Zen	38	Agents, orchestration, tool use	Temporary OpenCode promotion
3	mimo-v2.5	OpenCode Zen	37	Vision, long context, reasoning	Temporary OpenCode promotion
4	DeepSeek-V3.2	SambaNova	32	Reasoning and concise answers	Developer free tier
5	mistral-medium-3-5	Mistral	30	General agents and multimodal work	Free mode has evaluation limits
6	nemotron-3-super-120b-a12b:free	OpenRouter, Naga AI	25	Agent workflows with fast decoding	Explicit free model IDs
7	openai/gpt-oss-120b	Cerebras, Groq, Hugging Face	24	Fast reasoning and tool use	HF is credit-based; Cerebras/Groq are rate-limited
8	glm-4.7-flash	Z.ai	23	Bilingual reasoning, coding, agents	Token price is currently free
9	north-mini-code-free	OpenCode Zen	20	Focused coding tasks	Temporary OpenCode promotion
10	gemini-2.5-flash-lite	Gemini	11	Fast multimodal and large-context requests	Google free tier
10	gpt-4o	GitHub Models	11	Vision and general prototyping	Included quota; not intended for production
12	command-a-03-2025	Cohere	8	RAG, enterprise text, tool use	Evaluation key: 1,000 calls/month

The same model can appear through more than one provider. That is useful: adding both keys gives Hermes Router another quota pool and another endpoint to fail over to without changing model quality.

Provider speed and availability

Quality is only half of practical performance. Provider hardware, queueing, rate limits, and cold starts determine how quickly that model answers.

Provider	Practical advantage	Important limit
Cerebras	Extremely high output throughput for gpt-oss-120b	Free usage is rate and token limited
Groq	Fast gpt-oss-120b serving	Free-plan quotas vary by model
Gemini	Low latency and a 1M-token multimodal context	Flash-Lite favors speed over hard reasoning
OpenCode Zen	Access to several of the strongest free coding models	Free models are temporary and can rotate
OpenRouter / Naga	Two independent paths to Nemotron 3 Super	Shared free endpoints may queue under load
Hugging Face	One key can reach many inference providers	Free users receive only $0.10 monthly credit
GitHub Models	Convenient for prototypes using a GitHub token	Free API use is not intended for production

Hermes Router handles these differences through key rotation, health scoring, circuit breaking, and provider failover. For most apps, configure several providers instead of depending on the number-one model alone.

Use a ranked model

The built-in defaults already follow this list. To override one provider:

hr model set groq openai/gpt-oss-120b
hr model set zai glm-4.7-flash
hr model set opencode deepseek-v4-flash-free,nemotron-3-ultra-free,mimo-v2.5-free
hr restart

Check the active model and live health before relying on it:

hr model list
hr status

Sources and caveats

Availability was checked against current provider documentation: Gemini pricing, OpenCode Zen, Groq models and limits, Cerebras rate limits, Z.ai pricing, Cohere trial limits, GitHub Models free usage, and Hugging Face credits.

Free plans, promotions, model aliases, and benchmark results change. Treat this page as a starting point, then run your own prompts through the models that matter for your app.

---

Next: Configuration — override defaults and tune routing behavior.