OpenAI's GPT-6 Sol and GPT-6 Luna Launch on Amazon Bedrock, Priced Below GPT-5.6
OpenAI has launched GPT-6 Sol and GPT-6 Luna on Amazon Bedrock, positioned below flagship GPT-6 Astra for recurring coding tasks and high-volume document processing respectively. Both models cost less per API call than their GPT-5.6 predecessors, though exact pricing figures were not disclosed.
OpenAI's GPT-6 Sol and GPT-6 Luna are now generally available on Amazon Bedrock, filling out the middle and efficiency tiers of the GPT-6 model family beneath flagship model GPT-6 Astra.
The two models target different points on the intelligence-versus-efficiency curve. GPT-6 Sol is built for recurring, complex tasks — coding, debugging, refactoring, code review, data analysis, and multistep workflows across tools — where strong reasoning needs to run regularly rather than occasionally. GPT-6 Luna is designed for high-volume, focused tasks such as document extraction, summarization, classification, and targeted question answering, where latency and per-call cost compound across thousands of daily requests.
Both models are priced below their GPT-5.6 predecessors, according to OpenAI and Amazon, though neither company has published specific per-token rates. Pricing not yet disclosed in the source announcement.
Factuality and reasoning improvements
OpenAI claims GPT-6 Sol makes approximately half as many factual errors as GPT-5.6 Sol, based on an internal factuality evaluation — a company-reported figure not independently verified. OpenAI also says GPT-6 Sol communicates its work more clearly, flagging what it changed, what it verified, and what it could not confirm, which the company says helps developers identify gaps in automated task execution.
GPT-6 Luna reportedly received similar factual-reliability and communication improvements over its GPT-5.6 counterpart. It also supports adjustable reasoning effort per request, letting developers trade off quality, latency, and cost on a per-call basis.
Prompt caching and multi-model pipelines
Both models support explicit prompt caching on Bedrock, letting developers mark instructions, tool definitions, or reference material for reuse so repeated calls only process new input. Amazon and OpenAI position this as useful for coding assistants that repeatedly reference the same repository instructions, or document pipelines applying a consistent extraction schema.
The release also frames a three-tier model strategy: GPT-6 Luna for classification and triage, GPT-6 Sol for investigating complex cases, and GPT-6 Astra reserved for decisions where additional reasoning depth materially changes the outcome.
Infrastructure and data controls
On Amazon Bedrock, both models run with access governed through AWS IAM policies, invocation auditing via AWS CloudTrail, and VPC traffic isolation through AWS PrivateLink. AWS states inference runs on hardware-isolated infrastructure with zero-operator access, meaning AWS staff cannot view prompts or completions. Amazon says customer inference data is not used for model training and does not require opting into data sharing with OpenAI. Classifier-flagged traffic for abuse detection is retained up to 30 days; enterprises can request zero data retention through their AWS account team.
GPT-6 Sol and GPT-6 Luna are available now through the Amazon Bedrock console and supported APIs, with regional availability detailed in Amazon Bedrock documentation.
What this means
This release is less about a single model breakthrough and more about OpenAI and Amazon building a tiered pricing and capability ladder — Astra, Sol, Luna — that lets enterprises route workloads to the cheapest model that still meets quality bar, rather than defaulting everything to a flagship model. The lack of disclosed pricing, context window sizes, or standardized benchmark scores (MMLU, HumanEval, etc.) makes it difficult to independently assess how these models compare to GPT-5.6 or competing offerings from Anthropic and Google DeepMind. The claimed 2x reduction in factual errors is notable but unverified outside OpenAI's internal testing.
Related Articles
Google releases Gemini 4 Neon at $3/$15 per 1M tokens with a 1M-token context window
Google released Gemini 4 Neon, available today through the API at $3 per million input tokens and $15 per million output tokens. It is the fourth new Gemini 4 model in four weeks, with a context window of about one million tokens.
OpenAI Reports Up to 14% PPA Gain on AMD Flagship GPU IP From Model-Rewritten RTL
OpenAI reported on November 18, 2026 that its models achieved up to a 14% improvement in power, performance and area (PPA) on AMD flagship GPU intellectual property, measured against heavily optimized baselines. Some of the circuit rewrites were included in a design sent for manufacturing.
OpenAI models optimized circuits used in Jalapeño silicon and improved RISC-V core designs
OpenAI reports that its models optimized circuits used in Jalapeño silicon and improved RISC-V core designs. The company presents the work as progress in AI-assisted chip design with GPT-6 Astra.
KiteFishAI releases NanoI1, a 155-million-parameter open-weight model for local enterprise use
KiteFishAI has released NanoI1, an open-weight model with 155 million parameters. It targets specialized enterprise workflows on local hardware, according to Unite.AI.
Comments
Loading...