GPT-6 Astra

OpenAI · September 3, 2026

● activeClosedmixture of expertsmultimodal

Parameters~2.5T (MoE)

Context Window2M tokens

Why It Matters

Marked the official transition to the GPT-6 frontier generation, triggering industry-first critical-cyber safeguard frameworks.

Description

OpenAI's milestone frontier model marking the GPT-6 generation, and the first frontier system to trigger OpenAI's internal 'critical-cyber' safeguard threshold. Features continuous task-environment exploration and native tool synthesis.

Benchmark Scores

MMLUMassive Multitask Language Understanding — 57 subjects

93.8%

HumanEvalCode generation pass@1 — Python problems

96.4%

MATHMATH benchmark — competition-level problems

98.2%

GPQAGraduate-level science QA

82.5%

AIMEAMC/AIME math competition

98.4%

SWE-benchReal-world software engineering

74.2%

Key Innovations

Reasoning

ReasoningStructured step-by-step problem solving, often using chain-of-thought or tree-of-thought approaches.

Agentic

AgenticModels that can autonomously plan, execute multi-step tasks, use tools, and self-correct without human intervention.

Tool Use

Tool UseAbility to call external tools, APIs, and functions — enabling web browsing, code execution, and real-world actions.

MoE

MoEArchitecture where only a fraction of the model's parameters are active for each input, allowing massive scale with lower compute.

Long Context

Long ContextAbility to process very long inputs (100K+ tokens), enabling analysis of entire codebases or books.

Family Tree

Related Research (1)

Context Language Models (CLMs)Reasoning

2026 · arXiv:2609.37725

Introduced Context Language Models (CLMs) which dynamically read, append, and rewrite their active context window as an editable persistent file durin…