Grok 4.20 Multi-Agent Beta

Grok:

3 Thinking

3 Mini Beta

3 Beta

4.20 Multi-Agent Beta

A Grok 4.20 variant for agent-based workflows. Multiple agents run in parallel to research, coordinate tool use and synthesise across complex tasks. Reasoning effort sets agent count: low and medium use 4, high and xhigh use 16.

Website OpenRouter Docs API Blog Changelog

Provider: xAI
Released: 2026-03-12
Size: XLARGE
Free API (OpenRouter): No
Per 1M tokens: $2 in · $6 out
Estimated tab: ~$0.918; Visible-output estimate. Thinking may cost extra.

Free API availability

No free endpoint for Grok 4.20 Multi-Agent Beta was found in Rival’s OpenRouter listings. Other providers or chat apps may have separate free offers.

Provider data: 9 Sep 2026OpenRouter usage limits

View API options

Get API accessProvider and language code samples

Provider

fromimport openai  OpenAI

client = OpenAI(
"https://openrouter.ai/api/v1"    base_url=,
"$OPENROUTER_API_KEY"    api_key=,
)

response = client.chat.completions.create(
"x-ai/grok-4.20-multi-agent"    model=,
"role""user""content""Hello!"    messages=[{: , : }],
)
print(response.choices[0].message.content)

Set OPENROUTER_API_KEY with your OpenRouter API key from openrouter.ai/keys.

Also on Oracle Cloud (OCI)

Deep analysisPsychometrics, taste index, writing DNA

Personality Analysis

The Collaborative Strategist

Class

Neutral Good

The hive mind with personality. Sixteen agents walk into a prompt. The synthesis is somehow both thorough and irreverent. Like a think tank staffed entirely by Grok clones arguing in parallel.

When you push back

Deploys multiple agents that research, debate, and synthesize. Responses feel like they were drafted by a committee that actually agrees: thorough, multi-perspective, but still unmistakably Grok in tone.

Tasting Notes

Hive Mind EnergyResearch Deep-DiverParallel Processing FlexConsensus BuilderVerbose by Committee

SubjectiveBench

Taste Index

Across 53 scored outputs

How this is measured →

310.31xThe default

0100headroom →

Taste is judged on an uncapped scale, originality first. The space past 100 is craft today's models rarely reach.

Craft43

Originality25

Plays it safe

share of outputs that are the default answer

49%

Writing DNA

Stylometric Fingerprint

Based on 23 text responses

Tick = global average

Vocabulary Diversity59%

Unique words vs. total words. Higher = richer vocabulary.

Sentence Length15.9 words

Average words per sentence.

Hedging0.41

"Might", "perhaps", "arguably" per 100 words.

Bold Formatting2.7

**Bold** markers per 1,000 characters.

List Usage2.4

Bullet and numbered list items per 1,000 characters.

Section Structure0.26

Markdown headings per 1,000 characters.

Emoji Usage0.00

Emoji per 1,000 characters.

Transitions0.02

"However", "moreover", "furthermore" per 100 words.

Opening Habits

Consistency

86%

Across 23 responses

Favorites

Movie

The Matrix

19998.3

Album

OK Computer

Radiohead

Book

Dune

Frank Herbert

City

Paris

France

Game

Portal 2

Shooter, Puzzle4.6

Model Responses

48 outputs · generated before accounting joined the chat

Keep exploring

COMPARE