If §3.2 was the orchestrator/executor split and §3.3 was the empirical test, §3.4 is the third tier in the taxonomy — the auxiliary models. Auxiliary models are not for primary agent work; they handle web search, free grounding, free document processing, and the high-volume tasks the orchestrator and executor shouldn't be doing.

This article walks through the auxiliary picks in the channel's tier list, with a deep dive on Mimo V2 Pro (the high-volume king) and a routing rule for when to use the auxiliary slot vs when to keep the orchestrator and executor slots loaded.

What you'll learn

  • The auxiliary tier from the Top AI Models for Hermes Agent (Tier List) video: Gemini 2.5 Flash, Gemini 3 Flash, Mimo V2 Pro, Elephant Alpha, Trinity Large Preview.
  • Mimo V2 Pro's positioning as the "high-volume king" — the most-used model on OpenRouter for Hermes Agent users, free via the News API on the news portal website.
  • The Mimo V2 Pro deal: ~55% WildClaw success rate at $26 per suite (or free during the promotional window), official Hermes partnership, Xiaomi backing.
  • When to use Mimo vs Minimax: Mimo for high-volume and learning, Minimax for budget execution when Mimo's free window ends.
  • The self-evolving skill loop that makes Mimo a strong fit for Hermes Agent specifically: every task that completes with Mimo generates a reusable skill that survives model switches.

The auxiliary slot in the channel's tier list

The auxiliary slot is the third tier in the channel's model taxonomy. The models in this slot share three properties:

  1. Specialized capability. They win on a specific axis (web search, grounding, document processing) that the orchestrator and executor don't optimize for.
  2. Low cost. Many are free (Gemini 3 Flash, Mimo V2 Pro during the promotional window) or very cheap.
  3. Not for primary work. They fail on the general-purpose axes the orchestrator and executor win on.

The channel's auxiliary picks, in order of how often the channel routes to them:

  • Gemini 2.5 Flash — the default baked into Hermes; check nano config.yaml.
  • Gemini 3 Flash — adds free Google Search grounding and URL context reading, a replacement for the $10/month News Research built-in tools.
  • Mimo V2 Pro — high-volume king, the most-used model on OpenRouter for Hermes Agent users; free via the News API on the news portal website.
  • Elephant Alpha (100B params, 256K context) and Trinity Large Preview — open-weight niche picks.

The framing the channel uses: auxiliary models are for the tasks the orchestrator and executor shouldn't be doing. Don't route web search to GPT 5.4 when Gemini 3 Flash does it for free with grounding. Don't route document processing to Opus when Mimo V2 Pro does it cheaply (or free).

Gemini 2.5 Flash — the default in Hermes

Gemini 2.5 Flash is the default auxiliary model baked into Hermes Agent. The setup is in nano config.yaml — change the model id to swap in a different auxiliary. The model handles the basic support tasks (web search, grounding, simple document processing) at a low cost.

The case for Gemini 2.5 Flash: it's the default, it's cheap, and it's good enough for the auxiliary slot. If you don't have a specific reason to swap, leave it.

The case against Gemini 2.5 Flash: it's a step behind Gemini 3 Flash on grounding and URL reading. If your auxiliary workload leans on search results, upgrade to 3 Flash.

Gemini 3 Flash — free Google Search grounding

Gemini 3 Flash adds free Google Search grounding and URL context reading, a replacement for the $10/month News Research built-in tools. The implication: if your agent workflow needs search grounding, Gemini 3 Flash is the free alternative to the paid News Research tools.

The case for Gemini 3 Flash: free grounding, free URL context, and the Google Search quality is best-in-class. The replacement for the $10/month tools is real, not marketing.

The case against Gemini 3 Flash: it's still a Flash model, not a frontier model. Use it for search and grounding, not for general-purpose work. The channel's framing: "default auxiliary in Hermes Agent," with 3 Flash as the upgrade for search-heavy workflows.

Mimo V2 Pro — the high-volume king

Mimo V2 Pro is Xiaomi's AI model, specifically designed for agentic workflows and high-volume document processing. It's currently available for FREE through partner platforms, and it has quickly become the most-used model on OpenRouter for Hermes Agent users. Backed by Xiaomi's massive resources from phone and car sales, Mimo represents a serious entry into the AI model market.

The case for Mimo V2 Pro:

  • Currently FREE. Zero cost during the promotional period on News Portal, Kilo.ai (extended trial), and OpenRouter (limited free tier).
  • High-volume king. Most-used model on OpenRouter for Hermes Agent. Optimized for processing large documents, handles long agentic workflows efficiently, trained specifically for agentic use cases.
  • Official partnerships. News Research Team (creators of Hermes Agent) native integration, Xiaomi backing (phone manufacturer #2 in China, car manufacturer, long-term commitment to AI).
  • Agentic training. Tool calls integrate cleanly with Hermes' skill registry, native compatibility with agent frameworks, optimized for multi-step workflows.
  • Self-evolving skill loop. When you complete tasks with Mimo V2 Pro in Hermes Agent: it generates reusable skills automatically, skills carry over to future sessions, skills survive model switches, builds up agent capabilities over time.

The case against Mimo V2 Pro:

  • Free period is temporary. Will eventually require payment. The estimated price is $20–$40/month when the free period ends.
  • Less tested. Newer model with smaller community, limited English documentation.
  • 55% WildClaw success rate. Lower than GPT 5.4 (63–75%), higher than current Opus (40–51%). Real-world performance is solid but not frontier.
  • Unknown longevity. Will Xiaomi maintain commitment? The financial backing is real, but Xiaomi is new to the AI market compared to Anthropic, OpenAI, and Google.

Where to access Mimo V2 Pro

The free period is real, and the channel's recommendation is unambiguous: try it NOW while it's free. The access points:

  • News Portal (News Research Team website) — the official Hermes partnership route.
  • Kilo.ai (extended free trial) — the practical "use it in your own OpenClaw setup" route.
  • OpenRouter (limited free tier) — the broadest-access route, with the usual OpenRouter caveats.

The duration: "Free period is promotional and will end eventually." Plan your migration to Minimax M2.7 (the budget executor pick) when the free window closes.

The Xiaomi strategic context

Mimo V2 Pro is Xiaomi's entry into the agentic AI model market. The strategic context matters for tier-list building because Xiaomi is a fundamentally different kind of vendor than Anthropic or OpenAI:

  • Hardware giant. Xiaomi is a phone manufacturer (the #2 in China by market share) and a car manufacturer. The hardware revenue funds the AI research, which means the AI roadmap is not gated on a single funding round or a single product cycle.
  • Late entrant. Xiaomi is new to AI compared to Anthropic, OpenAI, and Google. The capability gap is real (Mimo V2 Pro at ~55% WildClaw is below GPT 5.4 at 63–75%), but the trajectory is positive. The next version will close part of the gap.
  • Chinese ecosystem. Xiaomi's primary market is China. The model is optimized for Chinese-language workflows, which is a real advantage for users with Chinese-language workloads.
  • News Research Team partnership. The official Hermes partnership means the integration is native. Xiaomi is committed to the agentic AI market, not just the chat market.

The take-away: Mimo V2 Pro is a credible bet for the auxiliary slot, and the Xiaomi backing means the model will keep improving. The "free" framing is a real cost advantage, but the strategic context is the long-term reason to consider Mimo over the alternatives.

The skill-generation mechanics

The self-evolving skill loop is the most interesting Mimo-specific feature for tier-list building. The mechanics:

  1. Task completion. You run a task with Mimo V2 Pro and Hermes Agent.
  2. Skill generation. After the task completes, Mimo analyzes the task structure and generates a reusable skill — a markdown file that documents the workflow.
  3. Skill storage. The skill is stored in Hermes's skill registry, not in Mimo's context.
  4. Skill reuse. On the next task, the skill is auto-loaded when relevant and the agent can invoke it.
  5. Cross-model portability. The skills survive model switches. A skill generated by Mimo works with GPT 5.4 or Minimax M2.7 the next time you swap.

The implication: the more you use Mimo with Hermes, the more capable the agent becomes — independent of which model is loaded. A skill generated by Mimo will work with any model that supports the Hermes skill format. The investment in Mimo is partly an investment in the Hermes skill library.

The channel's framing: "Investment carries forward." Build a skill library with Mimo while it's free, and the library keeps working when you migrate to a paid model. The skills are a Hermes artifact, not a Mimo artifact.

The Mimo vs Minimax decision

The most actionable routing rule in the auxiliary slot is the Mimo vs Minimax decision. The two models are close on capability (Mimo at ~55% WildClaw, Minimax at ~60–70% per the channel's tests) but very different on cost (Mimo free, Minimax $10–$20/month).

The channel's routing rule:

  • Use Mimo while it's free. No-brainer. Zero cost during the promotional period, official Hermes partnership, optimized for the workflows you're already running.
  • Use Minimax M2.7 when Mimo's free window ends. Same slot (executor / high-volume), better documentation, more mature community, predictable cost.
  • Don't become dependent on Mimo. The free period will end. Plan your migration before it does, not after.

The framing the channel uses: "Try it now, evaluate paid alternatives later." Mimo is the perfect "learn the workflow" model — you can experiment with high-volume document processing and agentic workflows at zero cost. When the free window ends, you have a clear sense of whether the workflow is worth paying for, and you can migrate to Minimax M2.7 or another budget executor.

The self-evolving skill loop

The most interesting Mimo-specific feature is the self-evolving skill loop. When you complete tasks with Mimo V2 Pro in Hermes Agent:

  • Mimo generates reusable skills automatically based on the completed task.
  • The skills carry over to future sessions.
  • The skills survive model switches (they're stored in Hermes's skill registry, not in Mimo's context).
  • The agent's capabilities build up over time as you complete more tasks.

The implication: the more you use Mimo with Hermes, the more capable the agent becomes — independent of which model is loaded. A skill generated by Mimo will work with GPT 5.4 or Minimax M2.7 the next time you swap. The skill is a Hermes-level artifact, not a Mimo-level artifact.

The channel's framing: "Investment carries forward." Build a skill library with Mimo while it's free, and the library keeps working when you migrate to a paid model.

The "Mimo V2 Pro is the perfect learning model" framing

For users new to the agent stack, Mimo V2 Pro is the perfect on-ramp:

  • It's free. Zero cost during the promotional period.
  • It's official. Hermes partnership means the integration is native, not bolted-on.
  • It's forgiving. 55% success rate is high enough to feel productive, low enough that failures teach you where the orchestrator/executor/auxiliary split matters.
  • It builds skills. The self-evolving skill loop means your early experiments leave artifacts you'll use forever.

The channel's framing: "Perfect for experimenting with AI agents." If you haven't built an agent stack before, Mimo is the cheapest way to learn. The skill library you build will outlive the free period.

The "build a skill library" recipe

The skill library is the most valuable artifact you can build during Mimo's free window. The recipe:

  1. Pick a recurring workflow. Email triage, code review, document summarization, game build. Pick the one you run most often.
  2. Run the workflow with Mimo. Let the agent complete the task and generate a skill.
  3. Review the skill. Read the generated markdown file. Confirm it captures the workflow accurately.
  4. Save the skill. The skill is auto-loaded by Hermes on relevant tasks. No manual saving required.
  5. Run the same workflow on a different model. Hot-swap to GPT 5.4 or Minimax M2.7 and run the same task. The skill should still work — the skill is a Hermes artifact, not a Mimo artifact.
  6. Iterate. Add custom skills, refine the generated ones, build a library.

The framing: the skill library is the long-term value of Mimo's free window. The model access is short-term; the skill library is permanent. Build the library while you can.

When NOT to use Mimo V2 Pro

The channel's "when to use Mimo" list is concrete:

  • Use Mimo if: it's free (no-brainer), you're learning, you have high-volume tasks, you're testing agentic workflows, you have zero budget, you're using Hermes Agent, the task is non-critical.
  • Avoid Mimo if: you need production-critical reliability, you can't wait for the free period to end, you need orchestrator-level intelligence (use GPT 5.4 or Gemini 3.1 Pro), you need a proven track record (use a model with a longer history).

The pattern: Mimo is the right pick for learning, testing, and high-volume non-critical work. It's the wrong pick for production-critical orchestrator work or any task that can't tolerate the eventual end of the free period.

The "production-critical" decision rule

The most actionable decision rule on Mimo is the production-critical distinction. The rule:

  • Use Mimo for non-critical work. Document processing, batch operations, data transformation, content generation at scale, learning workflows, testing agentic workflows. The 55% success rate is high enough to feel productive, low enough that failures are an acceptable cost.
  • Don't use Mimo for critical work. Production systems, mission-critical agents, anything that can't tolerate 45% failures, anything that needs orchestrator-level intelligence.

The framing the channel uses: "Non-critical work" is the sweet spot. Mimo is the right pick for the workflows where a 45% failure rate is a learning opportunity, not a crisis.

The "free period" planning rule

The free period is promotional, and it will end. The channel's planning rule:

  1. Use Mimo now. Don't wait. The free period is the cheapest time to learn the model.
  2. Document the workflows. Build a skill library while it's free. The skills survive the migration.
  3. Set a deadline for migration planning. When the free period end is announced, have a backup model configured and tested.
  4. Migrate gracefully. Don't migrate in a week. Test the new model on a few workflows before full migration. The channel's migration target is Minimax M2.7 at $10–$20/month.

The framing: "Try it now, evaluate paid alternatives later." The free period is a learning opportunity, not a permanent state. Plan the migration before the window closes.

The auxiliary slot vs the orchestrator and executor slots

The most common mistake the channel flags is routing auxiliary work to the orchestrator or executor models. Examples:

  • Routing web search to GPT 5.4. The orchestrator pays Opus-equivalent rates for a task Gemini 3 Flash does for free. Use the auxiliary slot.
  • Routing document processing to Opus. The executor pays Opus rates for a task Mimo V2 Pro does cheaply (or free). Use the auxiliary slot.
  • Routing search grounding to Minimax M2.7. The executor is optimized for tool calls, not search quality. Use Gemini 3 Flash.

The pattern: every model has a slot. Routing a task to the wrong slot costs more for the same output. The auxiliary slot is cheap because it's optimized for a narrow set of tasks — but the narrow set is exactly the work the orchestrator and executor shouldn't be doing.

The migration plan when Mimo's free window ends

The channel's recommended transition when Mimo's free window closes:

  1. Evaluate cost. Is paid Mimo worth it? The estimated price is $20–$40/month when the promotional period ends.
  2. Compare alternatives. For budget executor work, route to Minimax M2.7 ($10–$20/month) or DeepSeek V4 Flash (free on Nous Portal). For reliability, route to GPT 5.4 ($50–$75/month). For coding excellence, route to GLM 5.1 ($7–$10/month on Z.ai).
  3. Keep the skills. The skill library you built with Mimo works with any model. The skills are Hermes artifacts, not Mimo artifacts.
  4. Gradual switch. Test the new model on a few workflows before full migration. Don't migrate in a week.

The framing the channel uses: "Don't become dependent on Mimo." Build the skill library, learn the workflows, and have a backup model ready. When the free window ends, the migration is a config change, not a rewrite.

The "Mimo is not just a model" framing

The most under-discussed aspect of Mimo V2 Pro is that it's not just a model — it's a partnership. The Hermes Agent partnership is official, which means:

  • Native integration. The skill registry, the agent framework, the OpenClaw hooks — all native, not bolted-on.
  • Shared roadmap. The Hermes team and the Xiaomi AI team share a roadmap. Mimo's next version will be optimized for the Hermes agent framework.
  • Long-term commitment. The partnership is not a one-off. Xiaomi is committed to the agentic AI market, not just the chat market.
  • Cross-pollination. The skills generated by Mimo work with any model on Hermes. The features developed for Hermes (skill registry, agent framework, OpenClaw hooks) are used by Mimo natively. The ecosystem is the value.

The framing: Mimo's free window is a real cost advantage, but the partnership is the long-term reason to consider Mimo over the alternatives. A model with official integration beats a model with a community integration, all else equal. The channel's read: bet on the partnership, not just the model.

Try it yourself

  1. Sign up for free Mimo access via News Portal or Kilo.ai. The channel's recommendation is unambiguous: try it now while it's free.
  2. Test your high-volume workflows with Mimo. Document processing, batch operations, data transformation, content generation at scale — these are the workflows Mimo is optimized for.
  3. Build a skill library in Hermes Agent. Every task you complete with Mimo generates a reusable skill. The library outlives the free period.
  4. Compare Mimo with Minimax M2.7. Run the same task on both. The cost comparison is real (Mimo free vs Minimax $10–$20), and the capability comparison is the data point that decides your migration path.
  5. Try Gemini 3 Flash for search grounding. The free Google Search grounding is a real replacement for the $10/month News Research tools. If your agent workflow needs search results, Gemini 3 Flash is the auxiliary pick.
  6. Run the WildClaw benchmark on Mimo V2 Pro. The published score is ~55% at $26 per suite. Run it on your own machine to verify the score and add a custom test for your workload.
  7. Plan the migration. Don't wait for the free window to end. Have Minimax M2.7 (or another backup model) configured and tested. The migration is a config change, not a rewrite.
  8. Leverage the Hermes integration. Mimo is an official Hermes partner. The skill registry integration is native, not bolted-on. Take advantage of the official support.

Common pitfalls

  • Relying on Mimo for production-critical work. The 55% success rate is below frontier, and the free period is temporary. Use Mimo for non-critical work and have a backup model for production.
  • Routing orchestrator or executor work to the auxiliary slot. Auxiliary models are for specialized tasks — search, grounding, document processing. Don't route planning or tool calls to Mimo or Gemini 3 Flash.
  • Routing auxiliary work to the orchestrator or executor slots. The inverse mistake. Web search, document processing, and grounding should run on the auxiliary model, not the orchestrator. The cost savings compound.
  • Becoming dependent on Mimo's free window. Build the skill library, learn the workflows, and have a backup model ready. The migration is a config change, not a rewrite.
  • Paying full Opus rates for tasks that don't need it. Mimo V2 Pro and Gemini 3 Flash are the free alternatives for high-volume work. If your bill doesn't reflect the auxiliary slot, you're paying for capacity you don't need.
  • Betting on Mimo's longevity. Xiaomi is new to AI compared to Anthropic, OpenAI, and Google. The financial backing is real, but the long-term commitment isn't guaranteed. Plan for the model to be deprecated, not just for the free period to end.
  • Skipping the comparison with Minimax M2.7. The Mimo vs Minimax decision is the most actionable routing rule in the auxiliary slot. Run the same task on both, then decide.
  • Benching Mimo's 55% WildClaw score as a leaderboard ranking. The score is below GPT 5.4 (63–75%) and above current Opus (40–51%). Mimo is the right pick for high-volume and learning, not the right pick for frontier work.
  • Trusting "free" as a license to ignore rate limits. The free tier is rate-limited. The safe envelope for high-volume work is the WildClaw methodology, not unlimited free calls.
  • Treating the skill library as Mimo-specific. The skills are Hermes artifacts, not Mimo artifacts. They survive model switches. Build the library once, use it forever.

Sources

  • Top AI Models for Hermes Agent (Tier List) — 8,107 views · video_id: Af7Fg1m7hRw · cited: the auxiliary tier (Gemini 2.5 Flash, Gemini 3 Flash, Mimo V2 Pro, Elephant Alpha, Trinity Large Preview), the official Hermes partnership for Mimo
  • Minimax M2.7 is INSANELY GOOD! (Full Review) — 31,049 views · video_id: --uxieT5J9Y · cited: the Mimo vs Minimax comparison, the budget executor pick, the M2.7 trained-on-OpenClaw framing
  • Best Model for Openclaw (WildClaw Benchmarks!) — 4,574 views · video_id: 31Ij4Cum5tg · cited: the Mimo V2 ~55% WildClaw score, the $26 per suite cost, the 6-day free window via Kilo Code and partner providers
  • Supabase query — SELECT video_id, title, views, summary_content, summary_key_takeaways FROM public.videos WHERE video_id = ANY(ARRAY['Af7Fg1m7hRw','--uxieT5J9Y','31Ij4Cum5tg']); against project ttxdssgydwyurwwnjogq.