Same Day, Lower Prices: GPT-6 Sol and Claude Opus 5.5 Just Kicked Off an AI Price War
Image: Anthropic, official Claude Opus 5.5 announcement graphic (source).
On September 22, 2026, Anthropic and OpenAI both released new AI models, and the headline for both was price. Anthropic’s Claude Opus 5.5 costs 40% less to run than Opus 5 on typical workloads. OpenAI’s GPT-6 Sol and GPT-6 Luna cost half as much in the API as the GPT-5.6 models they replace. For anyone who pays for AI, whether through a subscription, a developer bill or a company budget, this is the most useful kind of news: you get more capability for less money. Here’s what changed, how the new models compare on price and what to do about it.
What Happened?
Anthropic went first. It introduced Claude Opus 5.5 as the first model in its new Claude 5.5 family. The company says it performs at the level of Claude Fable 5.1, its larger flagship model, on most work, while being much cheaper than Opus 5, which launched in July.
OpenAI followed with GPT-6 Sol and GPT-6 Luna. These are two smaller siblings of GPT-6 Astra, the flagship OpenAI launched earlier in September. According to TechCrunch, OpenAI’s announcement came about 90 minutes after Anthropic’s.
The timing makes the story bigger than two routine updates. Earlier this month, Anthropic publicly argued for pacing the frontier, meaning it wants to slow how quickly AI capabilities advance so that safety practices can keep up. Opus 5.5 is its first release since then. It is also a model positioned on efficiency rather than on raw capability, which fits that argument. As Fortune noted, the competition has shifted from building the most powerful model to delivering strong performance at a lower price.
The Key Details
Claude Opus 5.5 at a glance
- Price: $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. Tokens are the small chunks of text that AI usage is billed by.
- Cache reads: $0.20 per million tokens, 60% less than Opus 5. This is when the model reuses text it has already processed. Anthropic says cache reads make up most of the cost in agentic and coding work.
- Overall cost: Anthropic says Opus 5.5 costs 40% less than Opus 5 on typical workloads. That’s because it’s cheaper per token and uses fewer tokens per task.
- Speed: It generates output more than 30% faster than Opus 5.
- Writing style: Anthropic says it puts the most important information first, uses less jargon and follows your writing instructions more closely.
- Availability: It’s in Claude’s apps and API, plus Amazon Web Services, Google Cloud and Microsoft Azure. The API model ID is
claude-opus-5-5. - Subscription perks: Anthropic is raising five-hour usage limits on its Pro, Max, Team and seat-based Enterprise plans. It is also giving subscribers a one-time usage-limit reset they can save and use whenever they choose.
GPT-6 Sol and GPT-6 Luna at a glance
- GPT-6 Sol: $2 per million input tokens and $10 per million output tokens, down from $4 and $20 for GPT-5.6 Sol. OpenAI pitches it as the choice for harder tasks where you still want room to experiment.
- GPT-6 Luna: $0.10 per million input tokens and $0.50 per million output tokens, down from $0.20 and $1.20. It’s built for fast, high-volume work such as summarizing, extracting information and answering quick questions.
- Factual accuracy: On OpenAI’s internal factuality test, Sol makes about half as many mistakes as its predecessor.
- Caching: OpenAI says it has improved prompt caching for GPT-6, with a 90% discount when stored input is reused.
- Availability: Both models are in ChatGPT Work and Codex (OpenAI’s coding agent) for Plus, Pro, Business, Enterprise and Edu users. Free and Go users can use GPT-6 Luna in the ChatGPT desktop app. OpenAI says the models are “not yet available in Chat,” its regular chat mode. In the API they are
gpt-6-solandgpt-6-luna.
Image: OpenAI, official GPT-6 Sol and Luna announcement graphic (source).
Price comparison (per 1 million tokens)
| Model | Input | Output | Previous price (input / output) |
|---|---|---|---|
| Claude Opus 5.5 | $4.00 | $20.00 | Opus 5: $5.00 / $25.00 |
| GPT-6 Sol | $2.00 | $10.00 | GPT-5.6 Sol: $4.00 / $20.00 |
| GPT-6 Luna | $0.10 | $0.50 | GPT-5.6 Luna: $0.20 / $1.20 |
Prices from the official Anthropic and OpenAI announcements. OpenAI compares its new prices with GPT-5.6 “promotional pricing.”
Remember that price per token is only half of the cost. A model that finishes a task in fewer steps can cost less overall, even with a higher price per token. Both companies are leaning on that argument.
Why It Matters
Top-tier AI is getting cheaper fast
Opus 5 launched only two months ago. Its replacement is cheaper and, by Anthropic’s own measurements, better. OpenAI halved its API prices in a single release. When the leading labs cut prices this quickly, tasks that were too expensive to automate in the spring may be affordable now.
Both companies claim to beat the other, so treat the benchmarks carefully
Each announcement includes charts showing its model ahead of the other company’s. Anthropic says Opus 5.5 beats GPT-6 Astra on its FrontierCode coding benchmark at about 20% of the cost per task. OpenAI says GPT-6 Sol outperforms Claude Opus 5 on a business-workflow test at 9% of the cost per task.
Two points matter here. First, these are vendor-reported results, measured with each company’s own settings. Second, the comparisons often use the rival’s previous model. OpenAI’s charts compare against Opus 5, not Opus 5.5, because both launched the same day. Anthropic itself admits that “benchmark margins have become a less reliable guide to real-world differences” at this level. The honest conclusion: all three models are strong, and the only benchmark that settles it is your own workload.
Safety is now part of the sales pitch
Anthropic spends a large part of its announcement on safety. It says Opus 5.5 was evaluated before release by outside organizations, including METR and Frontier Design. It also says the model scored best of any model to date on Anthropic’s automated behavioral audit, a set of tests that checks how the model behaves across thousands of simulated scenarios.
The safety measures have practical effects for users. Because Opus 5.5 is highly capable at cybersecurity work, Anthropic says most cybersecurity tasks will be quietly handed to an older model, Opus 4.8. Routine bug-fixing in your own code stays with Opus 5.5. The model also can no longer be used with its “thinking” mode switched off.
OpenAI, for its part, says Sol and Luna show better alignment results than their GPT-5.6 versions. That includes fewer misleading claims about their own coding work.
What It Means for You
If you use ChatGPT or Claude day to day
- Free ChatGPT users can try GPT-6 Luna in the ChatGPT desktop app. It’s a good fit for summarizing documents, pulling key points out of long emails, or quick questions.
- Claude subscribers on Pro, Max, Team or seat-based Enterprise plans get higher five-hour limits. The saved usage-limit reset is handy for a deadline day when you’d otherwise run out.
- Expect shorter, clearer answers. Both companies say their new models waste fewer words and put the important information first.
If you're a developer
- Try the new model IDs (
claude-opus-5-5,gpt-6-sol,gpt-6-luna) in a test environment before switching production traffic. Don’t assume a same-day upgrade will behave exactly like the old model. - Take advantage of caching. Anthropic cut cache-read prices by 60%, and OpenAI offers a 90% discount on cached input. If your app sends the same long instructions or documents repeatedly, the savings can be large.
- Match the model to the task. Luna-class models are cheap enough for classifying, extracting or summarizing large volumes of text. Save Sol or Opus 5.5 for multi-step reasoning, coding and agent work.
- Note the policy changes. Opus 5.5 always runs with thinking on. Anthropic’s “preserved thinking” safeguard, which stops API users from editing Claude’s earlier reasoning in order to extract it, applies to API accounts created on or after August 31, 2026.
If you run a business or a startup
- Revisit your AI budget. If you set your 2027 AI budget using mid-2026 prices, recalculate it. Your cost per task may have dropped sharply.
- Test before you switch providers. Price cuts from both labs mean it’s worth comparing them on your own tasks instead of picking a side on loyalty or headlines.
- Plan for more agent work. Cheaper, faster models make long-running automated tasks, like code migrations, document processing and research reports, easier to justify financially.
What Happens Next?
Anthropic has confirmed that Claude Sonnet 5.5 and Claude Haiku 5.5 “will follow in the coming weeks” with many of the same improvements. Those are the cheaper, faster Claude models that many apps run on, so their pricing may matter even more to everyday developers than Opus 5.5 does.
OpenAI says GPT-6 Sol and Luna are rolling out gradually in ChatGPT and are not yet in regular Chat mode, so broader availability is likely the next step. The company has not given a date.
Beyond that, it’s reasonable to expect more price competition. Both launches explicitly compete on cost, and each company used the other’s models as its benchmark target. How that fits with both companies’ calls to pace AI development is an open question that deserves attention.
Final Takeaway
September 22 showed that the AI race now has a second scoreboard: price. Claude Opus 5.5 offers flagship-level performance at 40% lower cost than its predecessor. GPT-6 Sol and Luna halve OpenAI’s API prices. The practical advice is simple: try the new models on your real tasks, turn on caching, send simple work to the cheapest model that handles it well, and don’t pick a winner from the vendors’ own charts.
Sources: Anthropic, OpenAI, OpenAI Developer Community, TechCrunch (OpenAI), TechCrunch (Anthropic), Fortune, MacRumors. Benchmark results are as reported by each company and have not been independently verified by TheDevXpert.