AIToday
Large Language ModelsAI Business & IndustryArs Technica AIPublished: Sep 23, 2026, 10:00 JST

Anthropic Opus 5.5, OpenAI GPT-6 Sol and Luna: less cost

Anthropic Opus 5.5, OpenAI GPT-6 Sol and Luna: less cost

3 Key Points

  1. What happened

    Anthropic says Opus 5.5 saves closer to 40 percent versus Opus 5 for typical workloads at default settings, helped by fewer tokens used, and OpenAI says GPT-6 Sol and Luna cost half as much to use.

  2. Why it matters

    Both releases are pitched as modest upgrades that cut the cost of running frontier AI, so the comparison shoppers buying model access appear to be getting more capability per dollar.

  3. What to watch

    The OpenAI models are described as only sometimes a few percentage points better than their predecessors depending on the benchmark, so the value case hinges on which tasks a buyer actually runs.

WHO IT HITSProcurement and product teams choosing which frontier model to buy are the audience for these price claims, since both vendors are framing cheaper per-task costs as the main selling point.

Not sure about something? Ask the AI

Summaries like this, in your inbox every morning.

Context & Analysis

The two releases land in what the publisher calls the comparison shopping phase of the frontier AI race. Anthropic's Opus 5.5 carries the same protections that applied to Fable 5.1, so requests flagged as high risk in areas like cybersecurity and biology may be automatically and transparently routed to an older model. OpenAI's naming convention for its GPT-5.6 family — Sol, Terra and Luna — now extends to GPT-6, with Astra positioned as the most aggressively powerful and pricey model, and Sol and Luna as the cheaper options. The mapping between the two vendors' lines is imperfect, especially at the higher end. The stakes for buyers come down to how the cost claims hold up on the tasks they actually run, since the OpenAI gains are described as benchmark-dependent and only a few percentage points.

FAQ
How much cheaper is Anthropic's Opus 5.5?
Anthropic says the savings are closer to 40 percent compared to Opus 5 for typical workloads at default settings, helped by both lower token costs and fewer tokens used per task.
What is the difference between OpenAI's Sol and Luna?
Sol is a capable but more efficient and affordable alternative meant as a daily driver, while Luna is the fast, cheap option in the GPT-6 lineup.
Can GPT-6 Sol and Luna be used for anything GPT-6 Astra can do?
OpenAI says Sol and Luna were trained with similar methods to GPT-6 Astra, and depending on the benchmark they are sometimes a few percentage points more capable than their predecessors at certain tasks.
Ars Technica AIRead Original Article

Get the latest Large Language Models news every morning

For example, today's edition would include:

  • Identity Digital spins out Known Systems AISiliconANGLE AI · 1h ago
  • China probes Moonshot AI, DeepSeek over data leaksNikkei AI Stocks · 1h ago
  • Anthropic cuts Claude Opus 5.5 price 20%, OpenAI answers with cheaper Sol and LunaSiliconANGLE AI · 4h ago

AI-summarized, only the topics you pick — one digest a day via Email, LINE, or Slack.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. Q&As are published on this page for other readers too.

Related Articles

Next articleSnorkel AI valued at $3.5 billion in $350 million Series E