OpenAI and Anthropic introduced new models with lower published prices on Sept. 22, but the part buyers can verify is narrower than the launch narrative. OpenAI’s changelog confirms that GPT-6 Sol and GPT-6 Luna are live in the Responses and Chat Completions APIs, while Anthropic’s model page confirms Claude Opus 5.5 pricing and access terms. The performance claims matter, but they remain vendor claims unless a buyer tests them in their own workflow.
What OpenAI actually published
Reported facts. OpenAI says GPT-6 Sol and GPT-6 Luna accept text and image inputs and generate text through the Responses and Chat Completions APIs. For prompts with up to 272K input tokens, OpenAI lists standard prices of $2 input, $0.20 cached input and $10 output for Sol, and $0.10 input, $0.01 cached input and $0.50 output for Luna. OpenAI also points buyers to separate pricing for cache writes, longer prompts and other processing tiers.
Attributed claim. CNBC reported that OpenAI is positioning Sol for more complex work, including coding, and Luna for high-volume tasks such as extraction and summarization, while describing the launch as a 50% cut versus GPT-5.6 promotional pricing. That framing is important because it suggests the product split is not just about model quality; it is about matching different spend profiles to different workloads.
What Anthropic is saying about Opus 5.5
Reported facts. Anthropic says Claude Opus 5.5 is the first model in its Claude 5.5 family and costs 40% less to run than Opus 5. Its published list prices are $4 per million input tokens, $20 per million output tokens and $0.20 per million cache reads, compared with $5, $25 and $0.50 for Opus 5. Anthropic also offers a fast mode at $8 per million input tokens and $40 per million output tokens.
Attributed claim. Anthropic says the model was tested before release by external evaluators, and it presents its own benchmark table showing stronger scores than earlier models in areas such as coding, computer use and knowledge work. It also says Opus 5.5 is more resistant to prompt injection and less likely to take hard-to-reverse actions. Those are meaningful claims, but they are still self-reported and should not be treated as independent proof.
Anthropic’s pricing detail matters for a practical reason: the company says cache reads make up most agentic and coding costs. In other words, a model can look expensive on paper and still reduce spend if it reuses context efficiently, finishes tasks in fewer steps, or avoids expensive retries. That is the real mechanism behind many “lower cost” launches in frontier models: not just a cheaper sticker price, but less token churn per completed task.
What buyers should infer, and what they should not
Interpretation. The launches point in the same direction even though the product strategies differ. OpenAI is splitting its line into a higher-capability tier and an ultra-low-cost tier. Anthropic is keeping its top-end model expensive enough for serious workloads, but lowering the base price and promoting faster modes. For teams that run extraction, summarization, code review, tool-using agents or other high-volume jobs, those differences can affect unit economics quickly.
The boundary is equally important. None of the retrieved material independently proves that one model is better for a given buyer. OpenAI’s changelog confirms availability and prices, not comparative quality. Anthropic’s page offers detailed benchmarks and safety claims, but those are vendor-run evaluations with explicit caveats, and Anthropic itself notes that benchmark margins are less reliable guides at the frontier. If your use case is sensitive to latency, prompt caching, image inputs or long-context costs, those operational details may matter more than any single headline score.
For decision-makers, the immediate takeaway is not to chase the lowest list price alone. The useful question is whether your workload is dominated by input tokens, output tokens, cache reuse or latency, because that determines whether the advertised savings will actually show up in production. Check whether your workload is token-heavy, image-aware, or cache-friendly: those are the clearest signals that these launches can lower spend without proving better quality.
