
Aion 3.5 Review: AionLabs Roleplay Model, Pricing & When to Use
By Alex Morgan
MyClaw Editorial
MyClaw
Run Best-in-Class AI Agents Now
Run OpenClaw or Hermes Agent on managed MyClaw hosting, so your AI agent stays online, updated, and ready for real work.
AI Takeaway
- What is Aion 3.5?AionLabs’ GLM-based multi-model roleplay/storytelling system on OpenRouter (
aion-labs/aion-3.5), built for narrative tension — not general coding agents. - How does Aion 3.5 differ from Mini and 3.0?3.5 lifts context to ~262K vs 3.0’s ~131K at the same OpenRouter list stickers; Mini keeps that 262K window at a lower $/M for RP volume.
- What does OpenRouter list pricing look like?Moving list rates: 3.5 ~$3 / $6 per 1M in/out (cache ~$0.75); Mini ~$0.70 / $1.40 (cache ~$0.18). Marketplace tokens — not MyClaw invoices.
- Who should pick Aion 3.5 versus a general LLM?Choose 3.5 for two-character tension, long RP threads, or fiction pacing; skip it for multimodal jobs, tight coding agents, or buyers who need public benches.
- Can you run Aion 3.5 on MyClaw?No. Run Aion on OpenRouter or AionLabs; use MyClaw Chat or OpenClaw only for models MyClaw actually lists.
Aion 3.5 at a Glance
This review covers AionLabs’ specialist roleplay and storytelling stack — not the 2012 NCSoft Aion MMO patch, and not a desktop “Aion UI” shell. OpenRouter lists flagship aion-labs/aion-3.5 and sibling aion-labs/aion-3.5-mini. Cards describe a collaborative generation pipeline on the GLM family: several specialized models contribute to one reply so structure, tension, and conflict stay sharper than a single general chat model improvising a scene.
| Detail | Aion 3.5 | Aion 3.5 Mini | Aion 3.0 |
|---|---|---|---|
| OpenRouter ID | aion-labs/aion-3.5 | aion-labs/aion-3.5-mini | aion-labs/aion-3.0 |
| Context | ~262K | ~262K | ~131K |
| List $/1M in·out | ~$3 / $6 | ~$0.70 / $1.40 | ~$3 / $6 |
| Cache read (list) | ~$0.75 | ~$0.18 | ~$0.75 |
| Role | Full RP / story SKU | Lower-cost 3.5-class volume | Prior GLM-generation RP SKU |
Modalities stay text in → text out. The 3.5 card advertises tools / tool choice, JSON response_format without schema enforcement, and reasoning controls — useful when an RP harness needs structured side channels, not only freeform prose. Treat SillyTavern- or Janitor-heavy marketplace traffic as a usage signal, not a frontend endorsement. Older Aion-RP 1.0 (Llama-class character bench SKU) is a different product — do not mix IDs. For GLM-line context among Chinese labs, see the Chinese AI model roundup.
Aion 3.5 vs Mini vs 3.0: Which SKU Fits
Pick generation before price tier. Move to 3.5 when you want the ~262K window and the newer collaborative pipeline; stay on 3.0 only if a pinned harness still needs that ID and you accept ~131K. Inside 3.5, choose full 3.5 when tension quality beats unit cost, and Mini for long RP days, soft prompt tests, or many parallel threads on the same context class. Mini is not “discounted 3.0” — it is the cheap seat on the 3.5 band, so prompt packs and lore books should transfer with far less rewriting than a generation hop. Grabbing Aion-RP 1.0 is not an upgrade; that Llama-8B character SKU answers another bench. If the real job is a strong general Chinese LLM with tools — research briefs, coding agents, broad assistants — a peer such as the StepFun AI review is the wrong stand-in for Aion’s RP pitch, and a useful reminder that specialists and generalists shop differently.
OpenRouter List Pricing Without the Guesswork
OpenRouter’s public list (re-checked at optimize; stickers move) prices Aion 3.5 near $3 / $6 per million in/out, cache reads near $0.75. Mini lists near $0.70 / $1.40, cache near $0.18. Those stickers match 3.0 / 3.0-Mini even though context doubled — you pay for the specialist stack and window, not a quiet brochure hike. Infra snapshots (sub-second latency, ~50 tok/s class, high recent uptime) measure routing health, not prose quality. No independent public quality leaderboard existed for Aion 3.5 at launch, so do not budget from vendor adjectives alone. Cache discounts help long system prompts and lore books, but sticky output scenes still dominate the bill when replies run long. Marketplace $/tok is never a MyClaw Chat plan or AI Chat pricing line — MyClaw bills models and hosting it actually runs, and Aion is outside that inventory.
Honest Strengths and Limits
Pros:
- Purpose-built RP and storytelling with a collaborative multi-model pitch on GLM
- ~262K context on 3.5 and Mini — double 3.0’s window at the same list stickers
- Clear cost ladder: full 3.5 for peak scenes, Mini for same-generation volume
- Tools, JSON formatting, and reasoning knobs on the 3.5 card for structured RP harnesses
- Marketplace RP-frontend traffic shows where the product is actually used
Cons:
- No independent public quality benchmarks for 3.5 at launch — claims stay vendor- and usage-led
- Text-only — no native image or video path on these IDs
- Specialized: a weak default for coding agents, procurement, or general office loops
- 3.0 hands-on notes (huge ensembles / very long campaign state) are lineage caution only — not proven on 3.5
- Not available on MyClaw — no Chat soft-select and no assumed inventory arrival
If you need GLM-family generalist speed rather than an RP specialist, compare flagship-band options like GLM 5.3 Prime on their own terms.
When to Use Aion 3.5 (and When to Skip)
Use Aion 3.5 for character tension, betrayal beats, long multi-turn RP memory inside a ~262K window, or fiction pacing where “what happens next” beats tool correctness. Prefer full 3.5 for showcase scenes and reader-facing chapters; Mini for draft volume, lore expansion, and soft prompt iteration before you spend full-SKU tokens on a finale. Route those calls through OpenRouter or AionLabs.
Skip Aion 3.5 when you need screenshots or video in-loop, a tight coding or ops agent, third-party benches before purchase, or a huge ensemble that must stay consistent for dozens of sessions without extra state tooling. Prior 3.0 write-ups already flagged ensemble and ultra-long campaign drift as risk zones — treat that as lineage caution, not a measured 3.5 verdict. Pick a measurable general LLM instead, then host the loop where those IDs actually run.
If You Want Hosted Chat or Agents on MyClaw Instead

Aion stays off MyClaw. Use this path for hosted Chat or always-on agents on models MyClaw lists — writing, research, or general assistants — not a backdoor to aion-labs/aion-3.5.
Step 1: Open Chat and Pick a Listed Model
Open MyClaw AI Chat and choose a listed writing or general model. Skip Aion IDs — they are not on MyClaw. Guests get a limited try under automatic routing; signed-in plans unlock a fuller model list and history. Guest access is capped, not forever-free tokens.
Step 2: Freeze One Creative or Writing Ticket
Pin one scene, outline, or editorial ticket with a written pass/fail rule before you compare models. Score finished drafts and continuity — not vibes. If you still need Aion’s RP stack later, run it on OpenRouter beside MyClaw.
Step 3: Keep Winning Loops Online on OpenClaw
When prompts and tools should survive laptop sleep, move the MyClaw-supported loop into a managed OpenClaw agent. Hosting covers the always-on machine; Chat plans and token usage stay separate invoices — and still do not unlock Aion on MyClaw.
Final Call: Specialist RP Model Outside MyClaw Inventory
Aion 3.5 is a specialist RP and storytelling SKU: GLM collaborative generation, ~262K context, Mini for cheaper volume, and the same OpenRouter list stickers as 3.0 despite the larger window. Buy it on OpenRouter or AionLabs when narrative tension is the product and you can judge quality on your own scenes. Skip it — and do not expect MyClaw to host it — for multimodal inputs, coding agents, or public benches. On MyClaw, chat and host the models we actually carry; keep Aion on the marketplace that lists it.
Skip the Setup, Run Best-in-Class AI Agents Now
Launch a managed OpenClaw or Hermes Agent workspace in minutes, with always-on hosting, updates, and support handled by MyClaw.