Stay connected with KayaToday, follow us on Instagram and Facebook for the latest news and reviews delivered straight to you.
For the past two years, the central question in enterprise AI has been which model is most capable. That question is not going away, but a second one is now running alongside it with equal urgency: which model is cheapest to run at scale? OpenAI’s latest pricing move is a direct answer to that second question, and it signals that the frontier AI market has entered a new competitive phase where economics matter as much as intelligence benchmarks.
OpenAI announced cuts to two models in its GPT-5.6 series, reducing GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%, while introducing a premium Sol Fast mode priced at double the standard Sol rate. CEO Sam Altman described the changes on X simply as “major price cuts today.” The timing is pointed: the announcement came roughly a week and a half after Google launched its low-cost Gemini 3.6 Flash and Gemini 3.5 Flash-Lite models, and just days after Anthropic released Claude Opus 5 at the same sticker price as its predecessor Opus 4.8.
What the Luna cut actually changes in the market
Luna’s new pricing sits at $0.20 per million input tokens and $1.20 per million output tokens, giving a combined rate of $1.40 per million tokens. That is down from a combined $7 when the GPT-5.6 series launched in late June 2026 through a limited rollout initially made available by U.S. government request. The reduction is not cosmetic. It moves Luna from the middle of the commercial model market into a pricing tier previously occupied almost exclusively by smaller, specialist models from Chinese and open-weight providers.
To put that in concrete terms: Luna now costs less than Google’s Gemini 3.5 Flash-Lite, which carries a combined price of $2.80 per million tokens. It undercuts Alibaba Cloud’s Qwen3.7-Plus at $2.00 and Xiaomi’s MiMo-V2.5 at $2.40. The only models cheaper than Luna on a pure token basis are Xiaomi’s MiMo-V2.5 Flash at $0.40, DeepSeek’s flash model at $0.42, and a handful of promotional or limited-context offerings. Crucially, those cheaper alternatives do not carry the performance profile of an OpenAI frontier-series model. According to third-party benchmarking firm Artificial Analysis, Luna outperforms both Gemini 3.6 Flash and the older Gemini 3.1 Pro on its intelligence index, meaning the cost-per-unit-of-intelligence calculation now tilts meaningfully toward OpenAI for high-volume workloads.
AI coding startup Cognition made this point explicitly on X, noting that GPT-5.6 “sits on the pareto curve of price/performance efficiency,” illustrating how the series now offers among the strongest intelligence-to-cost ratios available commercially. That framing matters because enterprise buyers running coding agents, document processing pipelines, or automated classification systems do not simply compare token prices. They compare what a given budget actually produces in completed, useful work.
Terra and Sol tell a different story about OpenAI’s strategy
The Terra reduction is smaller in percentage terms but strategically significant. At a combined $14 per million tokens, Terra now matches Google’s Gemini 3.1 Pro Preview pricing for context windows up to 200,000 tokens. It also undercuts OpenAI’s own GPT-5.4, which remains at a combined $17.50. Krea AI’s Nic Dunz noted on X that Terra offers comparable intelligence to GPT-5.4 at roughly one-thirteenth the cost, a gap that makes the older model difficult to justify for most production use cases.
Sol moves in the opposite direction. The new Sol Fast mode, priced at $10 per million input tokens and $60 per million output tokens for a combined $70, is the most expensive configuration in the frontier model comparison. OpenAI says Fast mode delivers up to 2.5 times the throughput of Sol Standard without altering the model’s underlying intelligence. That is a deliberate premium positioning: rather than lowering Sol’s base price to compete with Anthropic’s Claude Opus 5, which costs a combined $30 per million tokens and performs at a comparable level, OpenAI is instead offering a speed tier for latency-sensitive enterprise workloads where throughput is the constraint rather than cost.
The three-tier structure that results is now sharply differentiated. Luna at $1.40 handles high-volume, low-complexity tasks such as summarization, routing, and lightweight real-time assistants. Terra at $14 serves general production use where capability and efficiency need to be balanced. Sol Standard at $35 and Sol Fast at $70 target the most demanding reasoning-heavy and agentic workloads, including multi-step planning, advanced coding, and tool-using systems.
Three companies, three theories of how to win on cost
What makes this moment interesting is that OpenAI, Google, and Anthropic are each pursuing a different answer to the same underlying question about how to lower the total cost of production AI work.
OpenAI is cutting per-token rates directly. Google is pairing lower prices with architectural efficiency, arguing that Gemini 3.6 Flash uses 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index, with savings reaching as high as 65% on some long-horizon engineering workloads. The implication is that even if Google’s per-token price is not always the lowest, the total token consumption across a task can be meaningfully reduced. Anthropic took a third path with Claude Opus 5: it held the price steady at $30 combined while replacing Opus 4.8 with a substantially more capable model, effectively lowering the price per unit of task completion rather than the price per token. Anthropic also added an adjustable effort setting that lets developers trade reasoning depth for speed and token savings.
None of these approaches is simply better than the others in all contexts. The right choice depends on the nature of the workload, the sensitivity to latency, and whether token volume or task complexity is the dominant cost driver. But the convergence of all three major frontier providers on cost as a primary competitive axis within the same two-week window is itself the signal worth watching.
Why this matters beyond the pricing table
For businesses in Malaysia and Singapore evaluating AI infrastructure, the practical implication is that frontier-grade language model capability is becoming substantially cheaper to access through APIs, and the gap between frontier models and cheaper alternatives is narrowing on the cost dimension even as it remains wide on the performance dimension. Developers building on OpenAI’s platform through Microsoft Azure or directly through the API will see Luna’s cost reduction reflected in their compute budgets almost immediately, which changes the economics of deploying AI at scale for tasks that previously required careful rationing of API calls.
The deeper shift is structural. Access to capable models is no longer the scarce resource it was even twelve months ago. The new constraint is operational cost predictability across large production workloads. OpenAI’s repositioning of the GPT-5.6 series, with Luna as its low-cost answer to the market’s cheapest tier, Terra as a direct challenger to Google’s pro-tier pricing, and Sol Fast as a premium throughput option, suggests the company understands that the next phase of enterprise AI adoption will be won not by the provider with the most impressive benchmark scores, but by the one whose models can run reliably and affordably at the volumes that real production systems demand.
Read More: AI agents write code easily but break on real pipelines. This framework is built to fix that
