GPT-5.6 Luna is the week’s quiet value winner

GPT-5.6 Luna is the clearest value story in this week’s market data. It pairs a quality rating of 85.3 with a $0.10 input price, producing a quality-per-dollar figure of 853. That is well ahead of every other model in the value table.

The result is easy to miss because there was no recorded price move. Luna simply appeared among the new models as a batch offering, with the same $0.10 input price and $0.60 output price. The cheapest listed Luna offer is even lower on input: $0.05 for batch use, with output at $0.30.

That makes OpenAI the quiet better buy this week—not because it cut an existing price, but because a strong model arrived at a price usually associated with much smaller systems. Its context window is 1050000, also the largest context figure in the new-model list.

The best quality for ordinary workloads

Luna’s quality score is 87, which puts it close to the top of the displayed value rankings without requiring premium-model pricing. For workloads where input volume dominates the bill, it is the first model I’d test.

The main caution is that the quality-per-dollar measure is not a complete workload bill. It is presented alongside input price, while output prices vary. Luna’s $0.60 output price is still low beside many high-end offerings, but buyers generating long answers should check both sides of the rate card.

DeepSeek V4 Flash 0423 is the other standout. Its input price is $0.14, quality is 80.8, and quality per dollar is 577.1. Its score of 86 is only one point below Luna’s 87. That combination gives it a credible claim for applications that don’t need Luna’s larger context window.

The data also shows why a simple “cheapest model wins” rule isn’t enough. Mistral Nemo is available from deepinfra at $0.019 input and $0.03 output, the lowest offer in the list. GPT-5 Nano is next at $0.025 input and $0.20 output. Those prices are attractive, but no quality rating is supplied for either model here, so I can’t rank them on quality per dollar.

The premium-value pick

GPT-5.6 Terra is the more convincing choice when quality matters more than the absolute lowest bill. It records quality of 91.1, the highest quality figure in the displayed value list, at a $1 input price. Its quality-per-dollar figure is 91.1, and its score is 90.

Terra also appears in the new-model list as a batch offering at $1 input and $6 output, with a context window of 1050000. That gives buyers a simple split: Luna for cost-sensitive, high-volume work; Terra for cases where the extra quality is worth a higher rate.

The rest of the value table is less compelling on the numbers supplied. GLM 5 has quality of 77 and quality per dollar of 77 at $1 input. GLM 5 Turbo reaches quality of 79.6 and quality per dollar of 66.3 at $1.20 input. Grok 4.20 posts quality of 75.1 and quality per dollar of 60.1 at $1.25 input.

A market with little price movement

This was a quiet week in one respect: the database records no price moves across 224 models and 125 providers. There were 12 new models, all OpenAI batch entries, while 8 models were removed. The removals included DeepSeek V4 Flash Latest, which is separate from the still-listed DeepSeek V4 Flash 0423.

What I’d actually do is start with GPT-5.6 Luna, compare it with DeepSeek V4 Flash 0423, and move to GPT-5.6 Terra only when the quality difference justifies the added cost. For an untested bargain, Mistral Nemo has the lowest listed price, but the available data doesn’t establish its quality.