If you are running batch workloads, the choice is straightforward: GPT-6 Luna is half the input price and 2.4x cheaper on output than the GPT-5.6-luna available on HiAPI, with a richer protocol and more reasoning controls.
GPT-6 Luna is live on HiAPI. Open the model page to try it directly, and new accounts get 200 free credits on sign-up.
One table says it all
| GPT-6 Luna | GPT-5.6 Luna | |
|---|---|---|
| Input ($/M token) | $0.025 | $0.05 |
| Output ($/M token) | $0.125 | $0.30 |
| Cached read ($/M token) | $0.0025 | $0.005 |
| Long context >272K input | $0.05 | $0.10 |
| Long context >272K output | $0.1875 | $0.45 |
| Context window | 1.05M | — |
| Max output tokens | 128K | — |
| Default protocol | Responses API | Chat Completions |
| Reasoning effort levels | Six (none/low/medium/high/xhigh/max) | Six (none through max, identical to 6-luna) |
| Disable reasoning to tune temp | Yes (reasoning.effort=none) | No |
Two numbers to remember: Luna input is half of 5.6-luna ($0.025 vs $0.05), and output is 2.4x cheaper ($0.125 vs $0.30). Long-context rates are halved too. Cached reads dropped from $0.005 to $0.0025, a difference that adds up fast at volume.
On the feature side: 6-luna defaults to the Responses API, adds none/xhigh/max effort levels, supports disabling reasoning so you can tune temperature and top_p, and increases the context window to 1.05M with a 128K output cap.
Same workloads, what is the real bill difference?
| Scenario | Monthly volume | GPT-6 Luna | GPT-5.6 Luna | Savings |
|---|---|---|---|---|
| Batch classification (input-heavy) | 100M in + 10M out | $3.75 | $8.00 | 53% |
| Content rewriting (output-heavy) | 10M in + 50M out | $6.50 | $15.50 | 58% |
| Conversational app (balanced) | 50M in + 50M out | $7.50 | $17.50 | 57% |
A typical batch classification pipeline at 100M input + 10M output tokens per month: $3.75 on 6-luna vs $8.00 on 5.6-luna. Send those same classification jobs through Batch processing for another 50% off, and you are at $1.88.
What is new beyond the price?
Responses API is now the default. The input array format, text.format with json_schema for structured output, and the reasoning configuration block are first-class on 6-luna. On 5.6-luna these are either unavailable or use different syntax.
Six effort levels instead of three. 5.6-luna shipped with low/medium/high. 6-luna adds none (disable reasoning and tune temperature), xhigh (heavy reasoning), and max (maximum effort). That means one model handles everything from zero-reasoning classification to deep analysis.
Tune temperature when reasoning is off. On 5.6-luna you cannot adjust temperature when reasoning is active. On 6-luna, set reasoning.effort=none and you get full control over temperature and top_p. Use the same model for both deterministic extraction and creative generation.
1.05M context window, 128K output. Feed entire manuals in one call. No more chunking long documents.
Using 5.6-luna now? Here is what you change.
Verdict: switch.
The price is half or less, and nothing is lost. The migration is a one-field change:
Chat Completions (no protocol change):
# Before
"model": "gpt-5.6-luna"
# After
"model": "gpt-6-luna"
Messages, temperature and all other fields stay the same.
Switch to Responses API (recommended for new code):
# Chat Completions (old path)
curl -s https://api.hiapi.ai/v1/chat/completions \
-H "Authorization: Bearer sk-YOUR_KEY" \
-d '{
"model": "gpt-6-luna",
"messages": [{"role": "user", "content": "Hello"}]
}'
# Responses API (new path, adds reasoning and text.format)
curl -s https://api.hiapi.ai/v1/responses \
-H "Authorization: Bearer sk-YOUR_KEY" \
-d '{
"model": "gpt-6-luna",
"input": [{"role": "user", "content": "Hello"}]
}'
Same API key, same domain. The only required changes: the model field and, if moving to Responses, the endpoint path and input format.
Are there any downsides to 6-luna?
None. 6-luna is strictly better on price, features and context window. If a pipeline on 5.6-luna depends on an edge-case Chat Completions behaviour, run a regression pass in Chat Completions mode on 6-luna before cutting over. For everything else, switch immediately.
FAQ
How much cheaper is 6-luna than 5.6-luna? Half the input price ($0.025 vs $0.05) and 2.4x cheaper output ($0.125 vs $0.30). Typical batch workloads save 53%–58% per month.
How much code do I change?
One field, model, if you keep the Chat Completions protocol. To use the Responses API you adapt the endpoint and request format, but the auth and parameter model stay the same.
Will 5.6-luna be removed? Both models are currently available on HiAPI. With 6-luna cheaper and more capable, there is no reason to stay on 5.6-luna.
What is the none effort level for?
It disables reasoning so you can adjust temperature and top_p. Use it for classification, extraction and formatting where reasoning overhead is unnecessary.
When does long-context pricing kick in? Automatically when a single request exceeds 272K input tokens. 6-luna has a 1.05M context window.
Want to see it in action? Open the model page, paste in a prompt you usually run on 5.6-luna, and compare the quality and cost. Or read the API guide to switch over in code.
