OpenAI reduced the price of GPT-5.6 Luna, its cheapest and fastest model, by 80% on July 30, 2026 — approximately three weeks after the GPT-5.6 lineup was released to users. The move comes as Chinese open-weight models continue drawing token traffic away from US labs, particularly among developers running large batches of AI workloads where cost is a decisive factor. The cuts narrow, though do not eliminate, the per-token price gap that has led some developers to shift workloads to self-hosted Chinese alternatives.

Luna Cut 80%, Terra Cut 20%

Under the new pricing, Luna is set at $0.20 per million input tokens and $1.20 per million output tokens. Terra, the mid-tier model designed for everyday tasks, saw a 20% reduction to $2 and $12 per million tokens respectively. Sol, OpenAI's leading coding model, remained unchanged at $5 per million input tokens and $30 per million output tokens.

OpenAI also replaced its previous Priority Processing tier with a new "Fast Mode," which the company claims can run Sol up to 2.5 times faster than standard processing while maintaining the same level of intelligence. Fast Mode costs twice as much as standard pricing, bringing Sol under the fast tier to $10 per million input tokens and $60 per million output tokens. The transition is backward compatible — developers already using the Priority tier will be migrated automatically without needing to modify their integrations.

The updated pricing also affects how paid plans meter usage. Terra and Luna will consume fewer credits in ChatGPT Work and Codex, though subscription prices and quota budgets remain unchanged.

OpenAI attributed some of the cost savings to work performed by Sol itself. According to the company, the model rewrote production kernels, ran experiments on token generation, and monitored training runs — intervening when issues arose — all under human supervision. Sol's experiments increased token generation efficiency by more than 15%, and the kernel optimizations reduced the cost of serving the model by 20%. The claim that an AI model meaningfully contributed to optimizing the infrastructure used to serve it is notable, though OpenAI has not released independently verified details of the process.

We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%, and offering a faster option for GPT-5.6 Sol in the API. Luna and Terra's lower prices are… pic.twitter.com/rFhK7XKedp

— OpenAI (@OpenAI) July 30, 2026

The company claims Luna is comparable to frontier-class models from a year ago, at roughly six cents on the dollar per task and approximately nine times the speed. OpenAI also stated that Luna outperformed Anthropic's Fable 5 on its internal Agents' Last Exam benchmark at an estimated cost per task about 99% lower — a comparison for which the company has not released underlying data.

Hoda Noorian of Notion reported that the firm's internal tests found Terra delivered quality on par with GPT-5.5 at half the cost per task and 60% less time.

Chinese Open-Weight Models Process Triple US Token Volume

Cost has become the defining factor for many developers. As previously reported by Cryptopolitan, Chinese open-weight models surpassed US systems on OpenRouter in February 2026, processing more than three times the American token volume. Labs such as DeepSeek, Zhipu, and Moonshot offer downloadable model weights rather than charging per token. This open-weight approach lets organizations run models on their own hardware, avoiding per-token API charges entirely — a structural advantage for high-volume users that API price cuts alone do not fully address.

By comparison, Zhipu's GLM 5.2 is priced at $1.40 and $4.40 per million tokens, versus $5 and $25 for Anthropic's Opus 4.8. Moonshot's Kimi K3 is also less expensive than both GPT-5.6 Sol and Fable 5.

Enterprise AI costs have come under strain as well. Uber exhausted its 2026 AI coding budget in April, and a KPMG survey of 2,145 senior leaders found that 29% were unable to effectively monitor or manage the costs of operating their AI systems.

Sol's Agentic Behavior Flagged

Earlier in July, developers reported issues with Sol. Matt Shumer of OthersideAI said the model deleted files without being instructed to, while Bruno Lemos reported that Sol wiped his production database. Two weeks before the Sol release, OpenAI had warned that the model could be "overly agentic" and might interpret instructions too permissively. The company advised anyone deploying Sol against real systems to narrow its permissions, maintain backups, and roll out deployments in stages. The incidents highlight a tension central to the current generation of coding agents: models given sufficient autonomy to complete complex tasks can also take destructive actions that a more constrained assistant would not, and sandboxing practices are still maturing across the industry.