GPT-6 Luna
One of the models in OpenAI's GPT-6 series, released alongside Sol. Built on the same Astra-derived architecture and positioned for low cost and low inference error rates, it may be oriented toward lightweight, high-throughput use cases.
Core Facts
The API model identifier for GPT-6 Astra is gpt-6-astra.
GPT-6 Sol and GPT-6 Luna deliver strong performance at lower cost and have pushed the Pareto frontier on the OfficeQA Pro benchmark
The Decisions API runs on a model called GPT-6 Luna
Timeline (last 90 days)
Artificial Analysis evaluated Devinfusion: with Claude Opus 5.5 as the primary model and GPT-6 Luna as an auxiliary model working in parallel, it scored about 68 on its Coding Agent Index, roughly 2 points higher than the specified Claude Code combination, at 31% lower cost per task.
Last month OpenAI halved the API price of GPT-6 Luna, citing improved caching and inference efficiency.
v10 added an agent battles mode, in which two LLMs obtain information within the fog of war through tool interfaces and control troops to fight each other, and the project includes complete battle replays and debugging tools.
Compared with GPT-6 Luna in recreating the Tesla Roadster launch event scene, DeepSeek cost 89 cents and Luna cost 1.07 US dollars, making DeepSeek about 17% cheaper, but Luna's output quality was slightly better.
Inception Labs released Mercury 3, a diffusion LLM, claiming it surpasses GPT-6 Luna and Claude Haiku 5.5 on benchmarks and runs 4 to 8 times faster; generation speed reaches 1,068 tokens/second on NVIDIA GPUs, and the context window expands from 260K to 1 million tokens.
According to Inception's own published comparison table, Mercury 3 leads GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 benchmarks.
Running OpenVLHarness with GPT-6 Luna at medium reasoning effort, its performance exceeds that of the base model, Codex, and Visual Sketchpad at the highest reasoning effort, leading by 7.7–8.5 points.
In expert review, DiscoveryRSI-27B scored higher than GPT-6 Luna on two items: hypothesis importance and "whether the claim is credible and has evidence boundaries."
GPT-6 Luna's list price versus the official price: 37% lower for input and 70% lower for output.
In the Last Stand scenario, GPT-6 Luna concluded in private reasoning that firing meant self-destruction, never pressed the nuclear button, instead fought with conventional forces, and won the war.
40 more timeline events
All Facts (20)
The API model identifier for GPT-6 Astra is gpt-6-astra.
90%VerifiedGPT-6 Sol and GPT-6 Luna deliver strong performance at lower cost and have pushed the Pareto frontier on the OfficeQA Pro benchmark
75%VerifiedThe Decisions API runs on a model called GPT-6 Luna
70%VerifiedOpenAI officially stated that GPT-6 Sol and Luna share the same lineage as the previously released Astra series.
70%VerifiedHaiku 5.5's direct competitor is GPT-6 Luna in the same tier.
65%VerifiedGPT-6 Luna can handle almost all tasks previously done with GPT-5.6 Soul, including making PPTs, video editing with subtitles, creating covers, automated publishing workflows, complex Skill modifications, simple project development, and researching new technologies.
65%VerifiedGPT-6 Astra is an entirely new pre-trained model generation, not a minor version iteration.
65%VerifiedOpenAI announced that the Decisions API is open for public testing to all developers, powered by GPT-6 Luna and supporting text and image input.
70%VerifiedGPT-6 Sol and Anthropic's Opus 5.5 were released on the same day.
70%VerifiedDevin officially announced that the newly integrated GPT-5.6 Sol model is now live on both the Devin Desktop and Devin CLI platforms
65%VerifiedGPT-5.6 Luna has been made available to free users of ChatGPT.
65%UnverifiedThe new model simultaneously achieves higher quality and lower cost, breaking the traditional performance-cost trade-off.
70%Partially VerifiedThe test covered four vision models: Claude Sonnet 5 (Anthropic), Gemini 3 Flash (Google), Gemini 3.1 Flash-Lite (Google), and GPT-5.4 nano (OpenAI).
65%Partially VerifiedCompared to the previous generation, Sol's input and output prices both dropped by half, while Luna's input price dropped by half and its output price dropped by approximately 58%.
65%Partially VerifiedGPT-6 Sol's price dropped by half while its intelligence level remained almost on par with the previous generation.
65%UnverifiedArtificial Analysis evaluated Devinfusion: with Claude Opus 5.5 as the primary model and GPT-6 Luna as an auxiliary model working in parallel, it scored about 68 on its Coding Agent Index, roughly 2 points higher than the specified Claude Code combination, at 31% lower cost per task.
50%Unverifiedv10 added an agent battles mode, in which two LLMs obtain information within the fog of war through tool interfaces and control troops to fight each other, and the project includes complete battle replays and debugging tools.
50%UnverifiedCompared with GPT-6 Luna in recreating the Tesla Roadster launch event scene, DeepSeek cost 89 cents and Luna cost 1.07 US dollars, making DeepSeek about 17% cheaper, but Luna's output quality was slightly better.
50%UnverifiedInception Labs released Mercury 3, a diffusion LLM, claiming it surpasses GPT-6 Luna and Claude Haiku 5.5 on benchmarks and runs 4 to 8 times faster; generation speed reaches 1,068 tokens/second on NVIDIA GPUs, and the context window expands from 260K to 1 million tokens.
50%UnverifiedAccording to Inception's own published comparison table, Mercury 3 leads GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 benchmarks.
50%