Anthropic and OpenAI released new models on the same day, and both are selling the same thing: work close to their flagship model at a much lower price. Anthropic’s Claude Opus 5.5 “performs at the level of Claude Fable 5.1 on most work,” the company says, while OpenAI trained GPT-6 Sol and GPT-6 Luna “with similar methods” as its flagship, GPT-6 Astra.
Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5, and reading from the prompt cache costs 60% less. Anthropic says the model also spends fewer tokens per task, so the total drops 40% on typical workloads at default settings, where effort is now medium. Moving code from Opus 5 takes more than a new model ID: the documentation lists four breaking changes, among them that thinking can no longer be switched off and that forcing a tool call now returns an error. Claude Sonnet 5.5 and Claude Haiku 5.5 are due in the coming weeks.
GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, half of GPT-5.6 Sol’s promotional price, though a prompt over 272,000 tokens is billed at twice the input rate and 1.5 times the output rate for the whole request. GPT-6 Luna costs $0.10 and $0.50. Both are available in the API and are rolling out to ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users; they are not yet in the ChatGPT chat. OpenAI’s lead number comes from AutomationBench, a test of business workflows across 47 tools: GPT-6 Sol at xhigh effort scores 33.2% at $0.27 per task, while Claude Opus 5 at max effort scores 26.9% at 11.1 times that cost. On an internal evaluation built from conversations where users flagged errors, Sol makes about half as many factual mistakes as its predecessor.
Neither company published a comparison with the other’s new model. Anthropic’s own table puts Opus 5.5 at 66.4% on Terminal-Bench 4.0 at xhigh effort, ahead of the 57.9% OpenAI reported for GPT-6 Astra, and two independent evaluators split on how much of that lead holds: Artificial Analysis, which ran Opus 5.5 with its default fallback on, measured 59.6%, level with Astra, and Vals AI 61.62%, 4.5 points ahead, or 53.54%, behind Astra, if the tasks its fallback models finished count as failures.
Two independent sources have tested both new models the same way, and both put Opus 5.5 ahead at a higher cost per task. On the Artificial Analysis Intelligence Index v4.3.2, at maximum effort, Opus 5.5 scores 58 points at $5.98 per task and GPT-6 Sol scores 48 at $1.06. On Zapier’s AutomationBench leaderboard, Opus 5.5 at max effort reaches 40.0% at $1.28 per task, against GPT-6 Sol’s best, at xhigh, of 33.2% at $0.27.
The Artificial Analysis data also qualify Anthropic’s saving. At maximum effort Opus 5.5 uses about 119,000 output tokens per task, against about 73,000 for Opus 5, which leaves its cost per task about level with Opus 5’s; at every lower effort setting it comes out cheaper than Opus 5 at the same setting.
By list price, GPT-6 Sol sits with Claude Sonnet 5 at $2 and $10, and Opus 5.5 sits below both companies’ flagships, GPT-6 Astra and Claude Fable 5.1, at $10 and $50. Both launches also promise plainer prose: Anthropic says Opus 5.5 is “less likely to use jargon or idiosyncratic phrases,” and OpenAI promises “more clarity, less jargon, fewer odd turns of phrase.”