OpenAI and Anthropic both cut the price of their working-tier frontier models on September 22, 2026. OpenAI released GPT-6 Sol and GPT-6 Luna at half the API price of the GPT-5.6 models they replace, and Anthropic released Claude Opus 5.5 at lower per-token rates than Opus 5. The same day, Snorkel AI, which sells training data to AI labs, announced a $350 million funding round on the strength of lab demand.
OpenAI halves prices on its mid-tier and small models
GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens in the API. GPT-6 Luna costs $0.10 and $0.50. Both are half the price of the GPT-5.6 Sol and Luna models, a drop OpenAI attributes to improvements in caching and inference (TechCrunch). Prices are as of September 22, 2026.
OpenAI positions Sol for complex work such as coding and Luna for clerical work. AWS, which made both models generally available on Amazon Bedrock on launch day, lists summarization, extraction, classification and routing as Luna's intended jobs, and says both models support up to 1 million tokens of context (AWS).
The price gap inside OpenAI's own lineup is now wide. GPT-6 Astra, the flagship released earlier this month, lists at $10 per million input tokens and $50 per million output (OpenAI). Sol is one fifth of that on both counts. OpenAI's main quality claim for Sol is narrower: on its internal factuality evaluation, built from real conversations where users flagged errors, it says Sol makes about half as many mistakes as its predecessor and reaches Astra-level reliability (OpenAI). That is OpenAI's own benchmark, and it has not been independently reproduced.
Distribution was broad from the first day. Sol and Luna went live in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users (OpenAI), and in GitHub Copilot (GitHub).
Anthropic cuts Opus list prices by a fifth
Claude Opus 5.5 lists at $4 per million input tokens and $20 per million output, down from $5 and $25 for Opus 5 (TechCrunch). Anthropic describes it as the first model in a new 5.5 family and says it performs at the level of Claude Fable 5.1 on most work (Anthropic).
The figure Anthropic leads with is larger than the list-price cut. It says Opus 5.5 costs 40% less to run than Opus 5 on typical workloads at default settings, and TechRepublic reports more than 30% faster output (TechRepublic). The per-token reduction is 20%. The rest of the claimed saving depends on how the model behaves on a given workload, so the 40% number is best treated as a vendor estimate until a team has measured it on its own traffic.
Opus 5.5 was available at launch through the Claude API and on Amazon Bedrock (AWS), and Anthropic lists Google Cloud and Microsoft Foundry as launch platforms as well.
What the two launches change
Both releases move capability down a price tier rather than raising the ceiling. OpenAI's case is that Sol matches Astra on factual reliability at a fifth of the price. Anthropic's case is that Opus 5.5 matches Fable 5.1 on most work below the old Opus price. Both are vendor claims resting on vendor benchmarks.
The rate cards are not claims. An organisation running a fixed workload on GPT-5.6 Sol or Opus 5 pays less per token this week than last, provided it migrates to the new model. Migration carries its own cost in regression testing, prompt adjustment and evaluation, and neither company's announcement removes that work. For teams whose AI spend was approved against last quarter's prices, the practical question is whether the savings justify a revalidation cycle now or at the next scheduled review.
Snorkel AI raises on the supply side
Snorkel AI announced a $350 million Series E at a $3.5 billion valuation on September 22, co-led by Insight Partners and S32 with participation from existing investor Addition (PR Newswire). TechCrunch reports that Lightspeed, Greylock, GV and Wells Fargo also took part, and that the valuation is nearly triple the $1.3 billion set in its $100 million Series D 17 months earlier (TechCrunch).
The company started as a data-labeling software vendor and shifted last year to delivering finished datasets, which it calls data-as-a-service. It produces them by combining its own software and model-generated data with subject-matter experts. Snorkel says that business has grown more than eighteenfold and crossed a $375 million annualized revenue run rate. That figure is company-reported and has not been independently verified.
One supplier's numbers are not a measure of the market. Still, Snorkel's round shows investors pricing in continued lab spending on training data in the same week OpenAI and Anthropic cut what they charge for finished models.
What to watch
Anthropic has called Opus 5.5 the first model in its 5.5 family and has not said when others will follow. Smaller 5.5 models would show whether its price reductions extend below the Opus tier. For OpenAI, the open question is whether the factuality result holds up in independent evaluations, since the case for moving work from Astra to Sol rests largely on it.
Sources
- OpenAI. "Introducing GPT-6 Sol and Luna." 2026. https://openai.com/index/introducing-gpt-6-sol-and-luna/
- OpenAI. "GPT-6 Astra Model." 2026. https://developers.openai.com/api/docs/models/gpt-6-astra
- TechCrunch. "OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes." 2026. https://techcrunch.com/2026/09/22/openai-launches-gpt-6-sol-and-luna/
- GitHub. "OpenAI's GPT-6 Sol and GPT-6 Luna now available." 2026. https://github.blog/changelog/2026-09-22-openais-gpt-6-sol-and-gpt-6-luna-now-available/
- AWS. "OpenAI GPT-6 Sol and GPT-6 Luna are now generally available on Amazon Bedrock." 2026. https://aws.amazon.com/about-aws/whats-new/2026/09/openai-gpt-6-sol-luna-on-amazon-bedrock/
- Anthropic. "Introducing Claude Opus 5.5." 2026. https://www.anthropic.com/claude-opus-5-5
- TechCrunch. "Anthropic releases Opus 5.5 with lower prices and Fable-level performance." 2026. https://techcrunch.com/2026/09/22/anthropic-releases-opus-5-5-with-lower-prices-and-fable-level-performance/
- TechRepublic. "Anthropic Launches Claude Opus 5.5 With Lower Prices and Faster Output." 2026. https://www.techrepublic.com/article/news-anthropic-claude-opus-5-5-pricing-performance/
- AWS. "Claude Opus 5.5 is now available on AWS." 2026. https://aws.amazon.com/about-aws/whats-new/2026/09/claude-opus-5-5-aws/
- Snorkel AI via PR Newswire. "Snorkel AI Raises $350M to Scale the Data Factory for Frontier AI." 2026. https://www.prnewswire.com/news-releases/snorkel-ai-raises-350m-to-scale-the-data-factory-for-frontier-ai-302886796.html
- TechCrunch. "Snorkel AI triples valuation to $3.5B as demand for AI training data booms." 2026. https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/