Aion 3.5: Specifications & Sources

Specification	Value and source
API model ID	aion-labs/aion-3.5Source
System type	Multi-model system built on the GLM family; several specialist models work together on each responseSource
Input modalities	TextSource
Output modalities	TextSource
Context window	262,144 tokensSource
Maximum output	32,768 tokensSource

Specification

Value and source

API model ID

aion-labs/aion-3.5Source

System type

Multi-model system built on the GLM family; several specialist models work together on each responseSource

Input modalities

TextSource

Output modalities

TextSource

Context window

262,144 tokensSource

Maximum output

32,768 tokensSource

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

EvalMap: English long role-play / Turns breaking no rule (Show measurement, test conditions, and source)

Source value: 69
Score: 69
Metric: Share of turns with no system-prompt rule broken
Unit: %
Category: EvalMap role-play system evaluation
Direction: Higher is better
Sample: 216
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: Russian long role-play / Turns breaking no rule (Show measurement, test conditions, and source)

Source value: 91
Score: 91
Metric: Share of turns with no system-prompt rule broken
Unit: %
Category: EvalMap role-play system evaluation
Direction: Higher is better
Sample: 216
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=Russian; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: English long role-play / Memory in play (Show measurement, test conditions, and source)

Source value: 0.99
Score: 0.99
Metric: Memory-in-play score (−1 to +1; AA-Omniscience-style)
Unit: rating
Category: EvalMap role-play system evaluation
Direction: Higher is better
Sample: 184
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: English long role-play / Hard questions at the end (Show measurement, test conditions, and source)

Source value: 0.97
Score: 0.97
Metric: Hard end-of-game question score (−1 to +1)
Unit: rating
Category: EvalMap role-play system evaluation
Direction: Higher is better
Sample: 78
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: English long role-play / Repeated four-word runs (Show measurement, test conditions, and source)

Source value: 0.027
Score: 0.027
Metric: Repeated four-word runs in narration (proportion)
Unit: ratio
Category: EvalMap role-play system evaluation
Direction: Lower is better
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: English long role-play / Distinct word pairs (Show measurement, test conditions, and source)

Source value: 0.923
Score: 0.923
Metric: Distinct word pairs in narration (proportion)
Unit: ratio
Category: EvalMap role-play system evaluation
Direction: Higher is better
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: English long role-play / Median seconds a turn (Show measurement, test conditions, and source)

Source value: 11.1
Score: 11.1
Metric: Median seconds per turn
Unit: seconds
Category: EvalMap role-play system evaluation
Direction: Lower is better
Sample: 216
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: English long role-play / Mean reasoning tokens a turn (Show measurement, test conditions, and source)

Source value: 178
Score: 178
Metric: Mean reasoning tokens per turn
Unit: count
Category: EvalMap role-play system evaluation
Direction: Lower is better
Sample: 216
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: English long role-play / Cost per 36-turn game (Show measurement, test conditions, and source)

Source value: 0.38
Score: 0.38
Metric: USD billed per 36-turn game
Unit: USD
Category: EvalMap role-play system evaluation
Direction: Lower is better
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=English; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: Russian long role-play / Memory in play (Show measurement, test conditions, and source)

Source value: 0.99
Score: 0.99
Metric: Memory-in-play score (−1 to +1; AA-Omniscience-style)
Unit: rating
Category: EvalMap role-play system evaluation
Direction: Higher is better
Sample: 170
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=Russian; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: Russian long role-play / Hard questions at the end (Show measurement, test conditions, and source)

Source value: 0.86
Score: 0.86
Metric: Hard end-of-game question score (−1 to +1)
Unit: rating
Category: EvalMap role-play system evaluation
Direction: Higher is better
Sample: 73
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=Russian; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

EvalMap: Russian long role-play / Repeated four-word runs (Show measurement, test conditions, and source)

Source value: 0.023
Score: 0.023
Metric: Repeated four-word runs in narration (proportion)
Unit: ratio
Category: EvalMap role-play system evaluation
Direction: Lower is better
Source type: independent-evaluator
Evaluator: EvalMap, Small tests
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Route=OpenRouter; model ID=openrouter/aion-labs/aion-3.5; language=Russian; run date=2026-09-25; six matched generated role-play scenarios, 36 turns each; up to 8,000 output tokens per turn; reasoning effort not stated.
Context: EvalMap measured the Aion multi-model role-play system through OpenRouter, not an isolated GLM checkpoint. Its six generated scenarios per language are a small sample, and the prompts and responses are not published. The evaluator says index differences of 0.02-0.03 are noise. The values describe this system run, not a general capability score.

Source EvalMap, Small tests

Aion 3.5

The response reached the token limit. No complete recording is available.

Reasoning (requested): low

Current attempt

Run date: Oct 11, 2026
API provider: AionLabs
Requested API model: aion-labs/aion-3.5
API model reported in response: aion-labs/aion-3.5
Reasoning setting: low
Pinned endpoint: AionLabs | aion-labs/aion-3.5-20260923
Temperature: Provider default
Token limit including reasoning: 8,192 tokens
Reasoning tokens: 5,585
Test protocol: community-visual-low-v1
Technical checks: Failed
Response time: 143.5 s

Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.

Type	Evidence and data date
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · Aion Labs model catalog (retrieved October 7, 2026) · Aion Labs model catalog · Editorial description reviewed October 7, 2026
Input	Documented modalities. Retrieved October 7, 2026. Source: Aion Labs API reference · Aion Labs API reference (retrieved October 7, 2026) · Aion Labs API reference · Editorial description reviewed October 7, 2026
Output	Documented modalities. Retrieved October 7, 2026. Source: Aion Labs API reference · Aion Labs API reference (retrieved October 7, 2026) · Aion Labs API reference · Editorial description reviewed October 7, 2026
Research status	Research date October 7, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources	standardized general-purpose benchmark scores Original wording The checked Aion catalog and API reference publish no scores. The located exact-model results are EvalMap's small custom role-play tests, not standardized cross-model benchmark suites; no exact Aion 3.5 result for a standard suite was located in this review.
Unresolved	base-checkpoint-specific scores Original wording Aion Labs describes Aion 3.5 as a multi-model GLM-family system. The checked sources do not identify its contributing checkpoints or attribute any score to one isolated checkpoint.
Additional source	EvalMap, Small tests (retrieved October 7, 2026)

Aion 3.5

Specifications and access

Measurements without matching peer values

Our tests

Pelican on a bicycle

Published prices

Sources and data date