Google: Gemma 4 26B A4B Instruct (MoE)
Google Gemma 4 26B A4B Instruct (MoE) API pricing
Gemma 4 26B A4B Instruct (MoE) costs $0.052 per million input tokens and $0.160 per million output tokens on Oxyy. These are the pay-as-you-go prices after the current 60% discount (Launch Promo; standard price $0.130 input and $0.400 output), and cached input tokens are billed at $0 per million. It accepts text, images and video and returns text, with a 262K-token context window. A cheaper Google option is Gemini 2.5 Flash-Lite at $0.040 / $0.160 per 1M tokens; for every model and rate, see Gemini API pricing.
Gemma 4's Mixture-of-Experts model. 25.2B total parameters but only 3.8B active per token, so it runs almost as fast as a 4B model while scoring close to the dense 31B.
Providers
Where this model runs, and what each route costs. Requests are sent to a healthy provider automatically; if one errors, the gateway retries against another serving the same model.
| Provider | Input /M | Output /M | Cache read /M | Latency | Throughput |
|---|---|---|---|---|---|
| $0.052 | $0.160 | $0 | 4.15s | 51 tps |
Capabilities (13)
Supported parameters (7)
A parameter not listed here may be removed before the request reaches the model. When it is, the request still succeeds and the response names it, with the reason, in provider_specific_fields.dropped_params. Thinking controls work as described below whether or not they are listed.
Thinking
Thinking can be turned off: send reasoning_effort: "none", or thinking: {type: "disabled"}.
Pricing
What this model costs on Oxyy, pay as you go. "Charged" is the rate a request is billed at; the standard price is shown beside it for reference.
| Rate | Standard price | Charged | Unit |
|---|---|---|---|
| Input | $0.130 | $0.052 | per 1M input tokens |
| Output | $0.400 | $0.160 | per 1M output tokens |
| Cached input | $0 | $0 | per 1M cached input tokens |
| Cache write (5 min) | $0 | $0 | per 1M cache-write tokens, 5-minute |
| Cache write (1 hour) | $0 | $0 | per 1M cache-write tokens, 1-hour |
| Cached audio input | $0 | $0 | per 1M cached audio tokens |
| Reasoning | $0 | $0 | per 1M reasoning tokens |
| Cache storage | $0 | $0 | per 1M tokens per hour stored |
Discount. Charged rates include the current "Launch Promo" discount of 60% on every rate.
Specifications
- Model ID
gemma-4-26b-a4b-it- Developer
- Type
- Text
- Input
- Text, images and video
- Output
- Text
- Context window
- 262,144 tokens
- Knowledge cutoff
- January 2025
- Released
- Apr 2, 2026
- Tokenizer
- Gemini
- Series
- Gemma 4
- Image input
- Up to 20 MB each
- Video input
- Up to 100 MB each
- Endpoint
/v1/chat/completions
Best for
Quickstart
Call gemma-4-26b-a4b-it with a POST to /v1/chat/completions on https://api.oxyy.ai, using your Oxyy API key from the OXYY_API_KEY environment variable. The same key works for every model in the catalog.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.oxyy.ai/v1",
api_key=os.environ["OXYY_API_KEY"],
)
response = client.chat.completions.create(
model="gemma-4-26b-a4b-it",
messages=[
{"role": "user", "content": "Explain unified AI APIs in one sentence."},
],
)
print(response.choices[0].message.content)Performance
Throughput is how fast the model writes (tokens per second — higher is better). Latency is total round-trip time (lower is better). TTFT is time-to-first-token — how long before you see anything appear (lower is better). Throughput and TTFT are measured on streaming requests, which are the only ones with a first-token moment to time.
Throughput
Latency
Time to first token
Cache hit rate
Activity
Token volume and request traffic to this model over time. Daily totals, UTC.
Prompt tokens measure input size. Reasoning tokens show internal thinking before a response. They are part of the output, so the completion bars show the rest of it and each day’s bar is its total.
Frequently asked questions
What is Gemma 4 26B A4B Instruct (MoE)?
How much does Gemma 4 26B A4B Instruct (MoE) cost?
What is the context length of Gemma 4 26B A4B Instruct (MoE)?
Does Gemma 4 26B A4B Instruct (MoE) support tool calling and structured outputs?
What inputs and outputs does Gemma 4 26B A4B Instruct (MoE) support?
Which API endpoint does Gemma 4 26B A4B Instruct (MoE) use?
What is the knowledge cutoff of Gemma 4 26B A4B Instruct (MoE)?
When was Gemma 4 26B A4B Instruct (MoE) released?
What other models does Google have?
Alternatives to Gemma 4 26B A4B Instruct (MoE)
Models of the same kind from other vendors, closest in price first. Prices are pay-as-you-go on Oxyy, through the same API key.