Summarize this article with:
-
OpenAI cut GPT-5.6 API prices on July 30, 2026, its second pricing move on the family since the tiers launched on July 9.
-
Luna drops 80%, from $1.00 and $6.00 per million input and output tokens to $0.20 and $1.20.
-
Terra drops 20%, from $2.50 and $15.00 to $2.00 and $12.00.
-
Sol is unchanged at $5.00 and $30.00, despite widespread reporting that framed the cuts as an across-the-board reduction.
-
OpenAI credits stack-wide optimisation, and says Sol rewrote production kernels itself, cutting serving costs around 20%.
OpenAI cut GPT-5.6 API prices on July 30, 2026. Luna fell 80% to $0.20 per million input tokens and $1.20 output. Terra fell 20% to $2.00 and $12.00. Sol, the flagship, kept its $5.00 and $30.00 pricing, and it helped pay for the cuts.
What Actually Changed
The cuts are not uniform across the family, which is the detail most summaries get wrong. Two of the three tiers moved, and the flagship did not.
| Model | Input before | Input now | Output before | Output now | Change |
|---|---|---|---|---|---|
| GPT-5.6 Luna | $1.00 /M | $0.20 /M | $6.00 /M | $1.20 /M | -80% |
| GPT-5.6 Terra | $2.50 /M | $2.00 /M | $15.00 /M | $12.00 /M | -20% |
| GPT-5.6 Sol | $5.00 /M | $5.00 /M | $30.00 /M | $30.00 /M | unchanged |
Two things are worth noting beyond the headline percentages. First, the 80% cut on Luna is a change of category, not a discount: at $0.20 per million input tokens, high-volume classification and extraction stop being a line item worth optimising. Second, all three tiers keep the same 1 million token context window, so the cheaper tiers did not lose capability to earn the lower price.
Why the Prices Fell
OpenAI attributes the reduction to optimisation across its serving stack rather than a margin decision. Three areas are named: improved hardware routing, better production inference software, and smarter context-caching algorithms.
The more interesting claim is about where those gains came from. OpenAI says part of the efficiency was generated by the models themselves, with Sol autonomously rewriting production code kernels to cut serving costs by roughly 20%.
That is why Sol matters here even though its price did not move. The flagship did not get cheaper to buy. It made the cheaper tiers possible, which is a different and arguably more significant claim about what frontier models are now used for internally.
What This Changes for Your Cost Model
A price cut on the cheap tiers does not simply reduce your bill. It moves the boundary at which routing a task to a cheaper model becomes worthwhile.
The tiering boundary moves down
Before the cut, Luna at $1.00 and $6.00 was cheap but not negligible, so batching and prompt trimming still mattered on high-volume tasks. At $0.20 and $1.20 the calculus changes: triage, intent detection, classification, and field extraction become close to free, and the work of optimising them stops paying for itself.
Caching compounds the saving
Cache reads are discounted 90% against the input price, and cache writes cost 1.25 times the standard input rate. On Luna that puts a cached read near $0.02 per million tokens. The break-even is low: reuse a prefix more than twice and caching is already cheaper than not caching.
Sol becomes a more deliberate choice
With a wider gap between tiers, sending everything to the flagship is harder to justify. The output-price spread between Luna and Sol is now 25 times. That is the number to hold in mind when deciding what genuinely needs frontier reasoning.
Practical Routing After the Cut
A tiered setup that reflects the new prices looks like this:
-
Luna for triage, classification, extraction, and formatting, where volume is high and each call is simple
-
Terra for standard generation, summarisation, and code review, which is where most production traffic belongs
-
Sol for multi-step reasoning, agentic tool use, and anything where a wrong answer costs more than the tokens
Here is a provider-agnostic call through one endpoint, so the tier is a parameter rather than a code path:
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["EDENAI_API_KEY"],
base_url="https://api.edenai.run/v3",
)
TIERS = {
"triage": "openai/gpt-5.6-luna", # $0.20 / $1.20 per 1M
"standard": "openai/gpt-5.6-terra", # $2.00 / $12.00 per 1M
"reasoning": "openai/gpt-5.6-sol", # $5.00 / $30.00 per 1M
}
def ask(task: str, prompt: str):
return client.chat.completions.create(
model=TIERS[task],
messages=[{"role": "user", "content": prompt}],
)
The Wider Point About Pricing Volatility
GPT-5.6 launched its API tiers on July 9, 2026, adjusted cache pricing on July 17, and cut two tiers on July 30. That is three pricing changes in three weeks on one model family.
Any architecture that hardcodes a provider SDK inherits that volatility as engineering work. Keeping the model as configuration behind a provider-agnostic gateway turns a price change into a config edit, and turns a competitor's cut into an option rather than a migration.
Conclusion
The headline is a price cut, but the accurate version is narrower and more useful: Luna is 80% cheaper, Terra is 20% cheaper, Sol costs exactly what it did before. The practical consequence is that the cheap tier is now cheap enough to absorb far more traffic, and the gap to the flagship is wide enough that routing deserves a deliberate decision rather than a default.
If you want to move traffic between tiers, or between providers, without touching application code, you can find them at Eden AI.
Login to the platform to test it yourself.
.jpg)
.png)