The news
OpenAI's GPT-6 Sol and GPT-6 Luna models are now available on Amazon Bedrock and in public preview on Snowflake Cortex AI, according to posts from Amazon (AMZN) cloud unit Amazon Web Services and Snowflake (SNOW) dated September 28, 2026. Both companies also announced new Anthropic Claude models. Microsoft (MSFT) had already made Sol and Luna generally available in its Microsoft Foundry service on September 22, 2026, alongside GPT-6 Astra, which Microsoft pitches at more demanding reasoning work.
In its weekly roundup, AWS said the two OpenAI models landed on Amazon Bedrock, its managed service for foundation models, over the preceding few days. AWS pitched Sol at heavy, ongoing engineering and operations work and Luna at narrower, repetitive jobs run in large volumes. It said both are priced significantly lower than their GPT-5.6 predecessors, without listing prices in the roundup.
The same AWS post said Anthropic's Claude Opus 5.5, which AWS called the first of the Claude 5.5 family, landed on Bedrock over the same few days. AWS said the model gets more done per token than Opus 5 and is aimed at agentic coding, meaning AI that writes and runs code across many steps, and at work that runs over long stretches.
Snowflake said GPT-6 Sol and Luna are in public preview through Cortex Inference, with support in Cortex AI Functions, Cortex Agents and its CoCo and CoWork products coming later. Snowflake said all calls run within its security and governance perimeter. It described Sol as the stronger choice for complex professional and technical work and Luna as geared to fast responses across large workloads.
In a separate post the same day, Snowflake said Claude Sonnet 5.5 is coming to Cortex AI in public preview, starting with Cortex Inference. Snowflake relayed two figures reported by Anthropic: a 70.6% score on Terminal-Bench 4.0, a test of how well AI agents complete tasks in a command-line environment, and output more than 30% faster than Sonnet 5.
Separately, in a September 28 opinion column for The Register, Corey Quinn argued AWS should embrace being a wholesaler of AI infrastructure rather than a builder of finished applications. He wrote that OpenAI started on Microsoft Azure and that AWS has had to catch up, and that OpenAI expanded its AWS commitment by $100 billion earlier this year.
The numbers
- Claude Sonnet 5.5 score on Terminal-Bench 4.0 (Anthropic, via Snowflake)
- 70.6%
- Sonnet 5.5 output speed vs Sonnet 5 (Anthropic, via Snowflake)
- More than 30% faster
- GPT-6 Sol and Luna pricing vs GPT-5.6 (AWS)
- Significantly lower; figures not given in the roundup
- OpenAI's expanded AWS commitment (Corey Quinn, The Register)
- $100 billion
Why CEOs should care
For CIOs and procurement leads, the practical change is choice without a new vendor contract. Teams on AWS or Snowflake can test OpenAI and Anthropic models side by side in environments their security teams may already have approved. Ask your account teams whether usage of these models draws down existing AWS or Snowflake spending commitments, how per-token prices compare with buying directly from OpenAI or Anthropic, and whether newer features reach the platform at the same time as the labs' own APIs.
For CFOs, availability of rival frontier models on the same platforms weakens the case for large, long minimum commitments to any single AI lab. Use the ability to switch as leverage in direct negotiations, and prefer shorter terms or volume tiers that do not lock you into one model family. Remember that Snowflake's access is a public preview, so check preview terms before moving production workloads.
For CISOs and data leaders, the pitch is governance. Snowflake says calls run inside its security and governance perimeter, and AWS offers the models through Bedrock, its own managed service. Ask each provider where prompts and outputs are processed, whether the model maker can see them, how long they are retained, and what logging you get for audits.
The bigger picture
In our view, frontier models are being sold more like wholesale goods. OpenAI's GPT-6 Sol and Luna now sit beside Anthropic's Claude Opus 5.5 on Bedrock, Snowflake is adding Claude Sonnet 5.5 next to the OpenAI pair, and the same OpenAI models are also on Microsoft Foundry. That may shift more of the customer relationship from the lab to the platform and push competition among labs toward price, efficiency and reliability, which is consistent with AWS highlighting lower prices for GPT-6 Sol and Luna and fewer tokens for Opus 5.5.
For AWS, Quinn's argument is that its strength lies in supplying parts to others rather than building finished applications. For Snowflake, adding recent models from both labs lets customers run AI work next to the data they already store there.
What’s next
Watch for Snowflake to move the models from preview to general availability and extend them to Cortex Agents and AI Functions, and for AWS to publish Bedrock pricing that buyers can compare against the labs' direct rates. Also watch whether GPT-6 Astra, which Microsoft already offers in Foundry alongside Sol and Luna, follows them onto Bedrock and Cortex AI.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
