Bottom line

Released on September 22, 2026, GPT-6 Luna is OpenAI's designated model for focused, high-volume tasks. Sporting a massive 1.05-million-token context window and a 128,000-token maximum output limit, it pairs sheer capacity with native support for tool calling, file search, web search, and computer use.

Signal

The standout signal is the structural shift toward immense context tightly integrated with active execution. OpenAI has confirmed that GPT-6 Luna supports a 1,050,000-token context window alongside a substantial 128,000-token output limit. Crucially, the model natively supports functions, web search, file search, and computer use out of the box. The emerging pattern is that OpenAI is positioning this as the high-efficiency workhorse of the GPT-6 tier, built to ingest massive inputs and orchestrate dense, agentic responses without hitting output ceilings.

Noise

Because Luna is positioned as OpenAI's "most efficient" model, operators might assume it is a stripped-down routing model. However, the confirmed feature set—specifically its capability for computer use and granular search integrations—implies a highly capable agentic engine, not just a lightweight summarizer. Furthermore, while telemetry from routing layers like OpenRouter shows early platform availability, operators should treat this as a moving launch-window snapshot and not conflate it with stable latency or throughput profiles.

Model profile

• Provider: OpenAI
• Release Date: September 22, 2026
• Context Window: 1,050,000 tokens
• Maximum Output: 128,000 tokens
• Key Capabilities: Functions, web search, file search, and computer use

Assessment

By prioritizing both extreme context length and high output, OpenAI has engineered a model explicitly tuned for intensive document processing, extensive code generation, and long-running autonomous tasks. The explicit inclusion of "computer use" as a supported feature signals a maturation in agentic workflows, pushing the model beyond passive text generation into active, environment-level manipulation.

Where it fits

GPT-6 Luna is built for scale. The operator read here is that Luna fits best in pipelines requiring massive ingestion—such as repository-scale code analysis, full-book editing, or sprawling legal discovery—coupled with the need to take action. It is an ideal fit for autonomous agent frameworks that need a high-efficiency brain capable of both long memory and active tool execution without the overhead of the heaviest flagship models.

Operator implications

The combination of a 1M+ context window and 128k output limit fundamentally alters pipeline architecture. Operators can now pass entire environments or document troves in a single prompt and rely on Luna to execute downstream tool commands or search functions. If the provider facts hold in production, the likely implication is a reduction in complex RAG (Retrieval-Augmented Generation) infrastructure for mid-sized enterprise datasets, as they can now comfortably fit entirely in-context while the model actively interfaces with systems via computer use.