Routing Architecture: Auto, Max & Eco
How ARMES dynamically routes prompts across Auto, Max, and Eco modes based on task complexity, reasoning depth, latency budgets, and context scale.
Intelligent Routing in ARMES
Choosing the right foundation model for every individual prompt is cognitively exhausting. Should you choose a dedicated reasoning engine for formal mathematics, an autonomous code anchor for full-stack architecture, an empathetic prose model for executive correspondence, or a high-throughput workhorse for a 300-page document ingestion?
ARMES eliminates model decision fatigue through three purpose-built routing modes:
The Three Routing Modes
1. Auto Mode (🧠): Adaptive Dynamic Routing
Default · Right-Sized Compute · Available on All Plans (Tier-Scaled Ceiling)
Auto Mode is the primary routing engine in ARMES, available across all subscription tiers (Free, Eco, Pro, and Ultra). It intelligently analyzes your prompt across four evaluation dimensions and dispatches it to the best-suited model from your tier's intelligence pool.
Auto Mode is designed to right-size compute and extend your spend:
- Workhorse & Apex Pairing: High-throughput, agile workhorse models absorb routine conversational, formatting, and extraction turns (~80% of volume). When multi-step reasoning, mathematical proof, or complex full-stack architecture is detected, the router automatically escalates to apex horsepower (~20% of volume).
- Tier-Scaled Intelligence Ceiling (
Ultra Auto > Pro Auto > Eco Auto > Free Auto): The pool of candidate models and the intelligence ceiling scale directly with your subscription tier—from economy models on Free (ARMES GPT FREE), to sparse MoE flagships on Eco (ARMES GPT ECO), to frontier anchors on Pro (ARMES GPT PRO), to unrestricted apex frontier models on Ultra (ARMES GPT ULTRA). An Ultra subscriber running in Auto Mode has a higher intelligence ceiling than a Pro, Eco, or Free subscriber on the same Auto setting. - Compute Spend Extension: By avoiding unnecessary frontier compute on routine turns, Auto Mode stretches your daily and monthly compute allowance while maintaining sub-second time-to-first-token.
2. Max Mode (🔥): Apex Reasoning Intelligence {#max-mode}
Uncapped Horsepower · Available on All Plans (Ultra Max > Pro Max > Eco Max > Free Max)
When tasks demand maximum reasoning rigor without compromise, Max Mode locks routing exclusively to the highest-intelligence reasoning engines in your subscription tier.
- Uncapped Reasoning Depth: Dedicated to formal mathematical proofs, complex multi-repository refactors, kernel-level debugging, and high-stakes strategic analysis.
- Apex Intelligence Guarantee: Bypasses cost-saving, latency, and token-conservation heuristics to route only to the highest-scoring foundation models in your tier's candidate pool.
- Tier-Scaled Apex Ceiling (
Ultra Max > Pro Max > Eco Max > Free Max): Max Mode is available on every agent across Free, Eco, Pro, and Ultra plans, and automatically uses your subscription tier for its routing logic and candidate pool. Free Max and Eco Max lock routing to the top reasoning engines in the Free and Eco MoE pools, Pro Max routes across the highest-ranked Pro-tier frontier anchors, and Ultra Max unlocks the highest ceiling on the platform by routing across Ultra-exclusive premium frontier models.
3. Eco Mode (🌿): High-Efficiency Intelligence
Everyday Intelligence · Minimal Energy Footprint · Unlocked on Eco, Pro & Ultra (Locked Upgrade CTA on Free)
Eco Mode directs queries exclusively to high-performance Mixture-of-Experts (MoE) and distilled architectures that activate only a small fraction (3%–6%) of their parameters per token.
- Intelligence per Watt: Delivers 92%–98% benchmark parity on everyday tasks with 10–30× lower compute and electrical energy consumption.
- Unified Across Plans & Agents: Every agent on ARMES displays Auto, Max, and Eco in its Mode Selector:
- The Eco Subscription ($9/mo —
ARMES GPT ECO): On the Eco subscription tier, Auto Mode (Eco Auto) and Eco Mode (🌿) use the exact same high-efficiency routing logic across the sparse MoE pool, while Max Mode (Eco Max) locks to the top reasoning models in the Eco pool. Including Auto, Max, and Eco onARMES GPT ECOunifies the Mode Selector across every agent and makes transitioning plans seamless. - The Versatile Eco Mode Toggle (
🌿on Pro & Ultra): On Pro and Ultra, subscribers can toggle Eco Mode on any agent for drafts, brainstorms, code reviews, and daily conversations to stretch their included compute budget 10–30× further, and switch back toAutoorMaxwhenever they want full frontier power. (On Free, Eco Mode appears in the Mode Selector with an upgrade badge.)
- The Eco Subscription ($9/mo —
- See Eco Mode Architecture for deep architectural details.
The Four Evaluation Dimensions
When Auto Mode evaluates an incoming prompt, it analyzes four core dimensions in less than 15 milliseconds:
1. Complexity, Consequence & Verifiability
- Verifiable Extremes: Tasks involving formal mathematics, theorem proving, closed-form derivations, or binary disassembly route to engines with dedicated test-time compute and recurrent reasoning depth.
- Substantive Implementation: Real-world full-stack development, database schemas, and commercial planning route to high-fidelity agentic anchors with verified low execution gaps on software engineering benchmarks.
2. Voice, Tone Nuance & Empathetic Cadence
- High-stakes creative writing, executive letters, personal counseling, and constructive code reviews require superior human cadence and candor.
- Dense mathematical models are guarded from these tasks; prompts with high emotional or editorial sensitivity route to editorial flagships with top-ranked human preference scores.
3. Latency & Throughput Budget
- Routine mechanical extractions, CSV/JSON transformations, quick conversational acknowledgments, and lightweight scripting do not require multi-minute deep inference.
- These queries route to high-speed execution workhorses operating at hundreds of tokens per second with sub-second time-to-first-token.
4. Context Window & Economic Cliff Guards
- When a prompt contains massive document dumps, multi-hundred-page regulatory filings, or large repository histories, routing protects user budgets.
- Models with steep extended-context pricing tiers are guarded; large corpora automatically route to native long-context architectures with 1M+ token windows.
Dynamic Operational Specialist Lanes
Rather than relying on static, hardcoded model assignments that rapidly become obsolete, ARMES organizes foundation models into empirical operational lanes evaluated by live benchmark performance:
Dynamic Model Registry
Because foundation models and benchmark rankings advance rapidly, ARMES maintains a dynamic, database-driven model registry rather than static documentation.
To review the live models, context windows, token pricing, and verified benchmark scores currently active in your subscription tier, visit the AI Models Directory or the AI Access Portal.
Inspecting Which Model Answered
ARMES maintains radical transparency. Automated routing is never a black box.
On every response from an agent:
- Look at the footer of the assistant's message bubble.
- The badge reveals the exact foundation model and provider laboratory that generated the response.
- Hovering or tapping the badge displays the token consumption, inference mode (
Auto,Max,Eco, or manual pin), and runtime telemetry.
Manual Model Pinning & Overrides
If you prefer to explore the catalog and lock in a specific model for a specialized task—or simply want to test how different labs respond:
- Manual Pinning (Available on All Plans — Free, Eco, Pro & Ultra): Open the Mode Selector in the top right of standalone Chat (
/chat) or the center-panel Chat Tab in Command Center (/command) to browse models by lab and manually pin any model unlocked in your tier's pool (Freepool →Eco + Freepool →Pro + Eco + Freepool → fullUltracatalog). See Manual Model Pinning & Selection for step-by-step instructions. - Max Mode (
🔥— Available on All Plans): Select the flame icon (🔥) in the Mode Selector to lock into the highest-intelligence reasoning models in your subscription tier for your most demanding tasks (Ultra Max > Pro Max > Eco Max > Free Max). - Eco Mode (
🌿— Eco, Pro & Ultra; Locked Upgrade CTA on Free): Toggle the leaf icon (🌿) in the Mode Selector to route through high-efficiency Mixture-of-Experts models for sustainable, low-compute sessions (on the Eco plan,Eco AutoandEco Modeshare the same high-efficiency routing logic). See Eco Mode Architecture. - Model Dossiers: To inspect full token pricing, context windows, and verified benchmark scores for every engine, explore the AI Models Directory.