Robots Atlas>ROBOTS ATLAS
1Test Model Full Fields
AI Models›GPT

1Test Model Full Fields

Alpha v1.0 · Family: GPT
Test AI model with every schema field populated — used to verify UI rendering across all surfaces.
⏳ Preview✓ Public access⚖ Open weights★ FeaturedLLMReasoning modelMultimodal📁 GPT
Context window
256K
tokens
Parameters
175B
parameters
Max output
16,384
tokens
Release date
1 May 2026
🏢OpenAIProducer · 2024🔬AnthropicResearch lab · 2024–2025💼MicrosoftInvestor
Access:APIHostedDownloadDeployment:☁ Cloud💻 Local📱 On-device

Overview

This document is a synthetic rendering test of the AI Model schema. Every available field is populated, including descriptive fields (PL + EN), references, benchmarks, pricing, and the security section. bold, italic and code.

H2 Section

  • First bullet item
  • Second bullet item
Sample test quote.
Classification
LLMReasoning modelMultimodal
Family: GPT
Applications
3D world generationTranslationHypothesis generation
Access & deployment
APIHostedDownload
CloudLocalOn-device
Weights: Open weights
Key parameters
📏 Context: 256K
🧩 Parameters: 175B
✓ Tools · ✓ Fine-tuning
📥 Input: text, image, audio, video…
Robotics
3D scene reconstructionBimanual manipulationDexterous manipulation
Platforms
Gemini Enterprise Agent PlatformAmazon SageMaker AI

Technical specification

Context window
256K
tokens
Parameters
175B
parameters
Max output tokens
16,384
tokens per response
Knowledge cutoff
31 Dec 2025
Knowledge boundary
License
Test License v1.0
Hardware requirements
Requires 8 × NVIDIA H100 80GB or equivalent GPU cluster with at least 640GB total VRAM.
Features:✓ Tool use✓ Fine-tuning
Modalities
⬇ Input
textimageaudiovideodocumentsstructured_datarobot_sensors
⬆ Output
textcodeimageaudiostructured_datarobot_actionsmotion_trajectories

Capabilities and applications

Native model capabilities
Function Calling
Category: planning
Streaming output
Category: reasoning
Interleaved Multimodal Input
Category: reasoning
Anomaly detection
Identifying observations that deviate from the normal pattern — key in fraud detection, machine monitoring, and security.
Category: other
Robotics
3D scene reconstructionBimanual manipulationDexterous manipulation
Application domains
3D world generationTranslationHypothesis generation

Benchmark results

3 benchmarks
GSM8K
accuracy · zero-shot, chain-of-thought
92.4%
📅 15 Apr 2026📄 Oficjalny technical report producenta (test)
Result reproduced across three independent evaluations (test).
DROP
pass@1 · few-shot (5-shot)
88.1%
📅 20 Mar 2026📄 Paper arXiv (test)
Test 5-shot.
TestBench-Custom
F1 · zero-shot
76.3points
📅 10 Feb 2026📄 Wewnętrzna ewaluacja (test)
Legacy benchmark without reference.

Pricing

Technical architecture

Deployment and security

🔒 Security / Enterprise
✓ Verified enterprise information

Pełne wsparcie enterprise: SOC 2 Type II, ISO 27001, GDPR-compliant, data residency w UE i USA, opcjonalne CMEK.

Wszystkie kontrole security dostępne w planie Enterprise. Audit logs retencja 90 dni standardowo, do 7 lat na żądanie.
Updated: 15 Apr 2026↗ Security documentation

Articles

Sources and related pages