OpenAI Launches GPT-6 Luna: Fast, Low-Cost Model With 1.1M Context Window

OpenAI has released GPT-6 Luna, the fast, cost-efficient model in its new GPT-6 series, according to a listing on OpenRouter. The model became available on September 22, 2026, and is positioned below GPT-6 Sol and GPT-6 Astra in OpenAI's tiered GPT-6 lineup.

Specs and Pricing

GPT-6 Luna ships with a 1.1 million token context window. Standard OpenAI routing prices the model at $0.10 per 1M input tokens and $0.50 per 1M output tokens, with a $0.010 per 1M cache-read rate. Reported performance for the standard endpoint is a 2.49-second P50 latency and 83 tokens per second throughput, with 100% uptime over the measured period.

Multiple serving configurations are available through OpenRouter:

• OpenAI (Standard): $0.10 / $0.50 per 1M, 2.49s latency, 83 tps
• OpenAI Flex: $0.05 / $0.25 per 1M, 1.33s latency, 30 tps
• Amazon Bedrock: $0.11 / $0.55 per 1M, 1.37s latency, 152 tps
• OpenAI Fast: $0.20 / $1.00 per 1M, 1.11s latency, 20 tps

A "GPT-6 Luna Pro" variant is also listed, which OpenAI describes as the same underlying model served with reasoning.mode set to "pro" for higher-quality responses on complex tasks, priced identically at $0.10 / $0.50 per 1M tokens in one listed configuration.

Positioning and Capabilities

According to OpenAI, GPT-6 Luna is built for high-volume, latency-sensitive workloads such as chat, classification, and lightweight agentic tasks. The company claims that at higher reasoning effort settings, Luna can handle complex software engineering and computer-use tasks that previously required the higher-tier Sol model. OpenAI also states that Luna shares the broader GPT-6 family's improvements in factual reliability and a clearer, more concise communication style.

No independent benchmark scores for GPT-6 Luna have been published or disclosed at this time. OpenAI has not released a technical report or specific benchmark comparisons for the GPT-6 family alongside this listing.

Luna fits into a four-tier GPT-6 structure alongside GPT-6 Sol (mid-tier, $2/$10 or $1/$5 depending on configuration) and GPT-6 Astra (flagship, $10/$50 or $5/$25 depending on configuration). All three tiers share the same 1.1M token context window, suggesting a shared underlying architecture with different compute allocations per tier rather than distinct context capabilities.

What This Means

GPT-6 Luna's pricing — as low as $0.05 per 1M input tokens under Flex routing — places it firmly in the budget tier for high-throughput applications like customer support bots, classification pipelines, and lightweight coding assistants. The multiple serving options (Standard, Flex, Bedrock, Fast) with different latency/throughput/cost tradeoffs give developers explicit control over the speed-versus-cost equation, a departure from single-endpoint model releases.

The claim that higher reasoning effort lets a "fast" tier model handle work previously requiring a higher tier is significant if verified independently, since it would let cost-sensitive teams access more capability without upgrading models. Until OpenAI or third parties publish benchmark data, this capability claim should be treated as unverified. The lack of published evaluation scores makes it difficult to assess how GPT-6 Luna compares to competing low-cost models from Anthropic, Google DeepMind, or open-weight alternatives.