model release

AionLabs Launches Aion 3.5, a Multi-Model Storytelling System Built on GLM

TL;DR

AionLabs has released Aion 3.5, a collaborative multi-model system for roleplaying and storytelling built on the GLM model family. It offers a 262K token context window at $3 per 1M input tokens and $6 per 1M output tokens.

3 min read
0

Aion 3.5 at a glance

Context window262K tokens
Input$3/1M tokens
Output$6/1M tokens

AionLabs has released Aion 3.5, the latest version of its purpose-built roleplaying and storytelling AI system, now available through OpenRouter. The model is built on the GLM family of models and extends the company's approach of using multiple specialized models in a single collaborative generation pipeline.

What's new

According to AionLabs, Aion 3.5 uses a "collaborative generation process" in which several specialized models each contribute to a single response. The company claims this produces stronger narrative structure and more compelling tension and conflict than single-model generation — a persistent weak point for general-purpose LLMs used in long-form fiction and character roleplay.

Aion 3.5 supports a 262,000-token context window, doubling the 131K context of its predecessor, Aion-3.0. Pricing is set at $3 per 1M input tokens and $6 per 1M output tokens, with cached input priced at $0.75 per 1M tokens — identical to Aion-3.0's rate card despite the larger context window.

On OpenRouter's infrastructure, the model has shown a median latency of 0.54 seconds and throughput of 53 tokens per second, with 100% uptime and availability over the past three days of tracked data. These figures reflect performance through AionLabs' own provider endpoint rather than independently verified benchmarks.

Alongside Aion 3.5, AionLabs also released Aion 3.5 Mini, a smaller, lower-cost variant using the same GLM-based collaborative architecture and matching 262K context window, priced at $0.70 per 1M input tokens and $1.40 per 1M output tokens.

Product lineage

Aion 3.5 is the sixth major release in AionLabs' Aion series. The company has iterated across different base model families: Aion-1.0 and Aion-1.0-Mini were built on DeepSeek-R1 with Tree of Thoughts and Mixture of Experts techniques for reasoning tasks; Aion-2.0 was built on DeepSeek V3.2; and Aion-3.0 shifted to the GLM family, the same lineage Aion 3.5 continues. A separate model, Aion-RP 1.0, is a fine-tuned Llama-3.1-8B base model the company says ranks highest in the character evaluation portion of the RPBench-Auto benchmark.

No independent third-party benchmark scores for Aion 3.5 itself were disclosed at release. AionLabs has not published a technical report detailing the specific GLM checkpoint used as the base model or training data cutoff.

What this means

Aion 3.5 targets a narrow but active niche: AI systems optimized specifically for roleplay and narrative fiction rather than general assistant tasks. The multi-model "collaborative generation" approach — routing pieces of a response through different specialized models — is an architectural bet that ensemble methods can outperform a single fine-tuned model on creative writing quality, though AionLabs has not published data isolating that claim from simply using a larger context window or better base model.

The pricing sits above budget-tier models but well below frontier general-purpose models like GPT-5 or Claude Opus, positioning Aion 3.5 as a specialized tool rather than a general competitor. Buyers evaluating it for production roleplay or storytelling applications should treat the uptime and latency figures as infrastructure-level metrics, not evidence of output quality, and look for independent benchmark comparisons before switching from established general-purpose models.

Related Articles

model release

Google releases Gemini 4 Neon at $3/$15 per 1M tokens with a 1M-token context window

Google released Gemini 4 Neon, available today through the API at $3 per million input tokens and $15 per million output tokens. It is the fourth new Gemini 4 model in four weeks, with a context window of about one million tokens.

model release

Thinking Machines Lab releases Inkling 2 Mini: 26B-parameter looped MoE with 3B active and 1M-token context

Thinking Machines Lab has released Inkling 2 Mini, a multimodal Mixture-of-Experts model with about 26B physical total parameters and 3B active per token. It repeats its middle layers three times, supports a 1M-token context, and ships with a DSpark drafter. Hugging Face reports it leads 32 of 48 benchmarks against four comparable models.

model release

NVIDIA releases Nemotron-Labs-Sigma: 14 open 3B and 8B continuous diffusion language model checkpoints

NVIDIA released Nemotron-Labs-Sigma, a set of 14 open continuous diffusion language model checkpoints at 3B and 8B parameters. According to MarkTechPost, Sigma denoises text in a learned 16-D embedding space, distills to few-step samplers, and decodes up to 4.20x faster than autoregressive decoding.

model release

Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license

Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.

Comments

Loading...