OpenAI Lists GPT-6 Luna Pro: A High-Reasoning Mode for Its Budget GPT-6 Model, Not a New Checkpoint

What was listed

OpenRouter has published a listing for GPT-6 Luna Pro, described by OpenAI as "the same underlying model as GPT-6 Luna, served with reasoning.mode set to pro for higher-quality responses on complex tasks." This is a configuration flag on top of an existing model, not a separately trained checkpoint. The listing shows a release date of September 22, 2026.

The numbers

Under OpenRouter's standard routing (OpenAI as provider), GPT-6 Luna Pro is priced at $0.10 per 1M input tokens and $0.50 per 1M output tokens, with a 1.1M token context window. Measured performance at time of listing: P50 latency of 3.03 seconds, throughput of 64 tokens/second, and 100% uptime from the OpenAI endpoint, against 96.85% OpenRouter availability over the trailing 24 hours.

OpenRouter also lists alternate serving tiers for the same underlying weights: a "Flex" tier at $0.05/$0.25 per 1M tokens (input/output) with no throughput data reported, and a "Fast" tier at $0.20/$1.00 per 1M tokens, also without throughput data. These represent different latency/cost tradeoffs for identical model weights rather than different models.

Where it fits in the GPT-6 lineup

GPT-6 Luna is positioned as the fast, cost-efficient tier of OpenAI's GPT-6 family, sitting below GPT-6 Sol ($1-2 input / $5-10 output per 1M tokens) and the flagship GPT-6 Astra ($5-10 input / $25-50 output per 1M tokens). According to OpenAI, at higher reasoning effort, Luna can handle software engineering and computer-use tasks that previously required the Sol tier. Each tier in the family — Luna, Sol, Astra — has both a standard and a "Pro" reasoning-mode variant, with Pro variants carrying higher output costs for the same context window and, according to OpenAI, higher-quality responses on complex tasks.

All GPT-6 tiers share a 1.1M token context window and are text-only models, distinct from OpenAI's separate GPT Image 2.5 and GPT Transcribe product lines also visible in OpenRouter's catalog.

What this means

GPT-6 Luna Pro is not a new model release in the traditional sense — it is a reasoning-effort setting layered on GPT-6 Luna's existing weights, exposed as its own listing for pricing and routing purposes. This mirrors a pattern already seen with GPT-5.6 Luna Pro and other "Pro" variants across OpenAI's catalog: rather than shipping a new checkpoint, OpenAI is exposing configurable inference-time compute (reasoning.mode) as a distinct SKU with its own price point.

For developers, the practical effect is a wider menu of cost/latency/quality tradeoffs within a single model family rather than a choice between genuinely different models. Teams optimizing for cost should note that the underlying weights for Luna, Luna Pro, and the Flex/Fast serving tiers are identical — the decision is purely about which reasoning effort and serving speed to pay for, not which model is smarter.