OpenAI Cuts GPT-5.6 Luna Price 80% Three Weeks After the Model Launched

Michele Catasta, Replit’s president and head of AI, supplied the testimonial OpenAI chose to publish with its new GPT-5.6 prices. Catasta wrote, “GPT-5.6 Luna is the closest we’ve come to intelligence too cheap to meter. I’ve never seen a model this affordable be this powerful, it’s unlocking use cases for Replit we didn’t expect to build for a long time.”

Luna’s API price fell 80 percent.

OpenAI made the cut on July 30, three weeks after launching GPT-5.6, Axios reported. The report noted that price cuts usually arrive months after a launch rather than three weeks and that cheaper Chinese open-weight models have pressed OpenAI and Anthropic to justify their higher costs.

What the cut actually costs

As of July 30, 2026, Luna costs $0.20 per million input tokens and $1.20 per million output tokens. The previous rates were $1 and $6. OpenAI also lowered Terra to $2 per million input tokens and $12 per million output tokens, from $2.50 and $15. Sol’s standard price did not change.

At the rates available July 30, Anthropic’s Claude Sonnet 4.6 cost $3 per million input tokens and $15 per million output tokens.

OpenAI’s July 29 post presented Luna as inexpensive at several reasoning settings on the Artificial Analysis Intelligence Index v4.1. At its maximum setting, the chart placed Luna at a score of 51.24 and an estimated $0.0571 per task. GLM-5.2 Max scored 51.09 at $0.2692 per task in the same presentation. The chart appeared inside OpenAI’s own post. The material released with the price change contains no comparison OpenAI did not itself present.

The July 30 ZeroHedge account cited Artificial Analysis’s cost-per-task index, where Moonshot AI’s Kimi K3 completed tasks for 94 cents against $1.04 for OpenAI’s Sol. Kimi K3 is a 2.8 trillion parameter open-weight model released this month, which customers can run on their own infrastructure, and it has beaten proprietary systems on benchmarks. DeepSeek V4 Pro ran at four cents per task, according to the account.

Where OpenAI says efficiency came from

In its July 29 post, OpenAI attributed the discounts to work across its models and serving systems. The company said Sol autonomously rewrote production kernels in a human-led process, reducing end-to-end serving costs by 20 percent. It also reported that experiments designed and run by Sol increased token-generation efficiency by more than 15 percent.

Those figures are OpenAI’s own accounting of its own systems. The company released no independent measurement of them alongside the price change. The same holds for OpenAI’s claim that Luna beats Fable 5 on Agents’ Last Exam at an estimated cost per task nearly 99 percent lower.

Fast mode replaced Priority Processing for Sol on July 30, running up to 2.5 times faster than Standard at twice the price.

The customer bill behind price pressure

Sam Altman had previously called costs “a huge issue,” Reuters reported. His comment came as enterprise buyers grew reluctant to approve large AI budgets without clearer returns and long-running agents consumed more tokens.

Telnyx had been spending roughly $100,000 a day on Anthropic, according to the documented case in the July 28 account by Forbes senior contributor Peter Cohan. After switching to Chinese startup Z.AI, the reported cost was about $100 per agent per day. The account as captured does not state how many agents Telnyx ran or whether the workloads and results were directly comparable.

Falling token prices squeeze expensive model providers, Cohan argues. His case identifies chipmakers and Chinese labs as the beneficiaries. Token costs fell from about $60 per million tokens in 2021 to roughly $0.06 by July 2026, according to figures he cites. By mid-2026, DeepSeek, Qwen and GLM had captured 46 percent market share, according to another figure in his account.

What the cuts leave unanswered

OpenAI entered this price cycle with an adjusted gross margin of about 33 percent in the first quarter of 2026, according to Control Plane’s account. The internal target was 46 percent. Revenue reached $5.7 billion in the quarter, and the operating loss was $3.7 billion. Inference costs were cited as the reason for the margin decline. The account said those costs quadrupled in 2025. OpenAI budgeted $32 billion for training during 2026.

The company has not said what the Luna and Terra reductions will do to that margin. Control Plane’s account put OpenAI’s first-quarter 2026 cash reserves at $73 billion and projected its full-year inference costs at roughly $14 billion. Control Plane placed those figures in the context of a potential public listing.

OpenAI filed a confidential draft S-1 with the Securities and Exchange Commission on June 8, 2026, after considering a September listing at a valuation of as much as $1 trillion. Advisers later counseled a delay to 2027, Reuters reported in late June.

Moonshot Releases Kimi K3 Weights as Amodei Rejects Open-Weight Ban

Moonshot AI released the weights for its Kimi K3 model on Monday, a system with 2.8 trillion parameters. Developers can now download, modify and self-host K3, which the report described as the world's

The Implicator

Anthropic's Opus 5 Cut Tokens 17% but Cost More to Benchmark Than Opus 4.8

Anthropic released Claude Opus 5 on July 24 with per-token API rates unchanged from Opus 4.8, the company announced. Artificial Analysis reported that its independent Intelligence Index run cost $3,83

The Implicator

Palo Alto Networks CEO Says AI Token Costs Must Fall Up to 90%

Palo Alto Networks CEO Nikesh Arora said on CNBC on Thursday that AI token costs need to fall as much as 90% to support large-scale enterprise adoption. He called OpenAI CEO Sam Altman’s claim that th

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai