Grok 4.20 Multi-Agent
grok-4-20-multi-agentgrok-4.20-multigrok-4.20-multi-agent- 🧠 Multi-agent variant of Grok 4.20 for collaborative deep research
- 🆕 Multiple agents run in parallel, then synthesize one answer
- 📏 2 million-token context window (catalog)
- 🔧 Native web search and tool orchestration
- 👁️ Accepts text and image inputs (catalog)
- 💬 Streams intermediate reasoning before final output
- 🏢 Built by xAI
- 🎯 Aimed at multi-step, well-sourced research workflows
xAI is an American artificial intelligence company and wholly owned subsidiary of SpaceX. The company develops AI systems under the Grok brand, spanning language models, image generation, and video synthesis. xAI has quickly established itself as a multimodal AI lab with…
Explore 8 more models by XAI (company) →Grok 4.20 Multi-Agent is a specialized variant of xAI's [[sibling:grok-4-20|Grok 4.20]] built for collaborative, agent-based workflows rather than single-pass inference. According to xAI's documentation, when a request is sent, multiple agents are launched to discuss and collaborate, each specializing in an aspect of the task — searching the web, analyzing data, or synthesizing findings — before producing a final response. The model can also run without built-in tools, with agents relying purely on their collective reasoning.
Compared with the standard Grok 4.20 it derives from, the multi-agent build adds an orchestration layer that coordinates parallel agents and cross-references information across sources, while retaining native web search and tool use. Per the catalog, it carries a 2 million-token context window and accepts both text and image inputs within that combined budget.
The model exposes reasoning traces during processing and streams its intermediate thinking before final output, per xAI's API examples. Within the broader Grok family, it sits alongside later flagship reasoning releases such as [[sibling:grok-4-3|Grok 4.3]], which targets agentic workflows and instruction following.
No independently verified benchmark scores for this multi-agent variant were available from primary or top-evaluator sources at the time of writing, so capability claims here are limited to documented features. It is offered through xAI's API as an on-demand model.
This About section is AI-generated from public sources via VeniceStats + Venice inference, with no human editing. It may contain inaccuracies.
| Seller | Reputation↓ | Routing | Input $/M | Cached $/M | Output $/M | Categories | API |
|---|---|---|---|---|---|---|---|
| Venice.ai Proxy 0x1f22…18c9 | 99 | #3 | $0.71 | $0.115 | $1.415 | chat,reasoning,vision,multimodal,web-search,x-search | openai-chat-completions |
| ▲ Apex Ant 0x73b4…e736 | 83 | #2 | $0.639 | $0.1266 | $1.2735 | chat,vision,multimodal,reasoning,long-context | openai-chat-completions |
| surplusintelligence.ai 0x0e49…8927 | 68 | #1 | $0.426 | $0.069 | $0.849 | anon,chat,cheap,frontier,multimodal,reasoning,research,tasks,vision,web-search,x-search | openai-chat-completions |
| antseed-neon-puma-944e 0x6650…944e | 3 | gated | $0.4842 | $0.23 | $0.965 | chat,math | openai-chat-completions |
| antseed-opal-badger-2580 0xc85d…2580 | 2 | gated | $0.71 | $0.71 | $1.415 | chat | openai-chat-completions |
| Apex TEE Test 0xe672…7955 | 0 | gated | $1.50 | $0.24 | $3.00 | chat,premium,vision,multimodal,reasoning,long-context | openai-chat-completions |
"Best price" and the seller table are live AntSeed catalog data (advertised $/1M tokens — or $ per generated image for unit-billed image models — not settled amounts). Reputation = on-chain trust (0-100). "Routing" = the SDK's default buyer routing (what the VPR desktop app ships with): a trust ≥ 60 gate on the effective reputation, then cheapest-first among routable sellers; live failover state (per-peer cooldowns) is buyer-side runtime and not included. Model knowledge (TLDR, provider, About) via the VeniceStats enrichment layer. Advertised catalog, not the model used in any specific purchase. "Usage on AntSeed" counts only settlements whose buyers share the per-model split on-chain (metadata v2/v3, opt-in), so every usage figure is a lower bound.