Gemma 4 31B is the largest dense model in Google's Apache 2.0 open-weights Gemma 4 family, released April 2026. it ships alongside the 26B A4B mixture-of-experts model, takes text and images, and has an optional thinking mode.
pick it if
want Apache 2.0 open weights you can also self-host
need image input and a thinking mode at a low flat rate
write long outputs, since output costs the same as input
skip it if
you need audio input; Google limits it to smaller Gemma 4 sizes
you rely on cached-prompt discounts to cut repeat-context cost
GPT 5.6 Luna was the lowest-priced tier of OpenAI's GPT-5.6 family, below Terra and Sol, which OpenAI says roughly matches the earlier nano tier. it retired on 2026-10-06, replaced by the cheaper GPT-6 Luna.
pick it if
you needed the cheapest GPT-5.6 tier for high-volume text and image work
you reused long prompts, since cached reads billed at a tenth of input
you wanted adjustable reasoning effort, from none up to max
skip it if
GPT-6 Luna is available: same window, half the input price
faq
which is cheaper, Gemma 4 31B or GPT 5.6 Luna?
Gemma 4 31B lists the lower output price; input prices match. on a coding-agent session (2M tokens in, 60,000 out) Gemma 4 31B costs $0.41 and GPT 5.6 Luna $0.47.
which has the larger context window?
GPT 5.6 Luna offers the larger context window. Gemma 4 31B: 128k tokens; GPT 5.6 Luna: 1M tokens.
try either model
switch between models inside the same octomind session.