Gemma 4 26B A4B (free) API
Gemma 4 26B A4B is a sparse Mixture-of-Experts model from Google DeepMind with 25.2B total parameters and ~3.8B active per token, accepting text and image inputs.
Gemma 4 26B A4B is an open-weight generative AI model in the Gemma 4 family, developed by Google DeepMind. It uses a Mixture-of-Experts (MoE) architecture with approximately 25.2 billion total parameters, but only about 3.8 billion are activated for each token during generation. The model accepts both text and image inputs and produces text output. It includes native function calling, native system prompts, and a configurable thinking mode for reasoning. A dedicated draft model supports multi-token prediction and speculative decoding. The model is distributed via Kaggle, Hugging Face, and NVIDIA NGC, and is available as a fully managed, serverless offering on Google Cloud's Model Garden. An instruction-tuned version, Gemma 4 26B A4B IT, is optimized for dialogue and instruction-following tasks, while the base version suits developers who need further fine-tuning.
In terms of performance, Gemma 4 26B A4B reaches 97% of the 31B dense model's performance while significantly reducing inference compute requirements. It scores 82.6% on MMLU Pro, compared to 85.2% for the 31B dense model. On AIME 2026, it is only 0.9 percentage points lower than the 31B model. The model shows notable improvements on coding benchmarks. Thanks to its efficiency, it can run on devices such as a 32GB M2 MacBook Pro and supports multiple precision formats, including default 16-bit and lower precision via quantization.
This model is well-suited for high-throughput scenarios, advanced reasoning tasks, agentic workflows, local and edge computing, coding assistance and software development, language learning and scientific research, multimodal tasks, and building autonomous agent systems. Some users report that it may be slightly behind certain competitors, such as Qwen 35B, on specific coding tasks.
Updated:
MODEL ID
gemma-4-26b-a4b-it:free- Context
- 262.144K
- Max output
- 32.768K
- Recommended API
- messages
- Input
- image / text / video
- Output
- text
Highlights
- MoE architecture with 25.2B total and 3.8B active parameters
- Reaches 97% of 31B dense model performance
- Supports text and image inputs
- Configurable thinking mode for reasoning
- Native function calling and system prompts
- Runs on 32GB M2 MacBook Pro
Providers
Compare API prices across providers. Select a provider to view all charges and conditions.
| $0 | $0 |
Benchmarks
- 82.6%
MMLU Pro
31B dense model scores 85.2%
- 仅低0.9个百分点
AIME 2026
compared to 31B model
Scores as published by the model provider.
Reasoning
- Enabled by default
- No
- Always enabled
- No
Quick Start
Drop-in code to call this model with Felo API Platform compatible APIs. Start in the Playground when you want to test prompts before writing code.
Get your API key
Create a key from your Felo API Platform account and set it as an environment variable.
export FELO_API_KEY="YOUR_API_KEY"Make your first request
Use this model ID with the Felo API Platform endpoint. Optional request headers such as X-Request-Id can help troubleshooting.
gemma-4-26b-a4b-it:freeThis model is reasoning-capable. If your selected protocol and model route expose reasoning controls, preserve the returned reasoning metadata when continuing a conversation.
const response = await fetch("https://openapi.felo.ai/api/v1/messages", {
method: 'POST',
headers: {
Authorization: `Bearer ${process.env.FELO_API_KEY}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
"model": "gemma-4-26b-a4b-it:free",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}),
});
if (!response.ok) throw new Error(await response.text());
console.log(await response.json());Using third-party SDKs
Use compatible SDKs by setting baseURL to https://openapi.felo.ai/api/v1. Use the Anthropic-compatible base URL for Claude-style clients. https://openapi.felo.ai/api.
Enable streaming
Add stream: true to receive responses as server-sent events.
curl -N https://openapi.felo.ai/api/v1/messages \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemma-4-26b-a4b-it:free",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
],
"stream": true
}'Endpoint
Felo API Platform exposes compatible model-calling surfaces. Pick the surface your app or agent client already supports.
Messages
Anthropic-compatible messages for Claude-style clients.
POST
https://openapi.felo.ai/api/v1/messagesParameters
modelstringgemma-4-26b-a4b-it:freePublic model ID to call.max_tokensinteger—Maximum generated tokens.messagesarray—Conversation messages for Chat Completions.include_reasoning———reasoning———response_formatobject—Request structured output when supported.seed———temperaturenumber—Controls response variety when supported.tool_choicestring or object—Controls whether a tool should be called.toolsarray—Tool calling definitions following compatible schema.top_p———Frequently asked questions
What is Gemma 4 26B A4B (free)?
Gemma 4 26B A4B is an open-weight generative AI model in the Gemma 4 family, developed by Google DeepMind. It uses a Mixture-of-Experts (MoE) architecture with approximately 25.2 billion total parameters, but only about 3.8 billion are activated for each token during generation. The model accepts both text and image inputs and produces text output. It includes native function calling, native system prompts, and a configurable thinking mode for reasoning. A dedicated draft model supports multi-token prediction and speculative decoding. The model is distributed via Kaggle, Hugging Face, and NVIDIA NGC, and is available as a fully managed, serverless offering on Google Cloud's Model Garden. An instruction-tuned version, Gemma 4 26B A4B IT, is optimized for dialogue and instruction-following tasks, while the base version suits developers who need further fine-tuning. In terms of performance, Gemma 4 26B A4B reaches 97% of the 31B dense model's performance while significantly reducing inference compute requirements. It scores 82.6% on MMLU Pro, compared to 85.2% for the 31B dense model. On AIME 2026, it is only 0.9 percentage points lower than the 31B model. The model shows notable improvements on coding benchmarks. Thanks to its efficiency, it can run on devices such as a 32GB M2 MacBook Pro and supports multiple precision formats, including default 16-bit and lower precision via quantization. This model is well-suited for high-throughput scenarios, advanced reasoning tasks, agentic workflows, local and edge computing, coding assistance and software development, language learning and scientific research, multimodal tasks, and building autonomous agent systems. Some users report that it may be slightly behind certain competitors, such as Qwen 35B, on specific coding tasks.
How is usage billed?
Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices.
What input modalities does Gemma 4 26B A4B support?
It accepts text and image inputs and generates text output.
What is the context length of Gemma 4 26B A4B?
The model supports a context length of 256K tokens.
What is the difference between Gemma 4 26B A4B and Gemma 4 26B A4B IT?
The IT version is instruction-tuned for dialogue and instruction-following, while the base version is better suited for developers who want to fine-tune the model for specific needs.
How does Gemma 4 26B A4B compare to the 31B dense model?
It achieves 97% of the 31B dense model's performance while requiring much less inference compute. On MMLU Pro it scores 82.6% versus 85.2%, and on AIME 2026 it is only 0.9 percentage points lower.
Can Gemma 4 26B A4B run locally?
Yes, it can run on devices like a 32GB M2 MacBook Pro and supports multiple precision formats, including 16-bit and lower precision via quantization.
Related models
Gemini 3.8 Flash
Googlegemini-3.8-flashGoogle's Gemini 3.8 Flash is a fast, cost-efficient workhorse model for agentic coding, multimodal understanding and long-horizon tool use, with a 1M-token context window.
Context
1.049M
Max output
65.536K
Input tokens
$1.5 / 1M tokens
Output tokens
$7.5 / 1M tokens
Gemini 4 Argon
Googlegemini-4-argonGoogle's flagship Gemini 4 model with a unified 1M-token context window and up to 1M-token output, built for long-horizon software engineering, enterprise knowledge work, and cyber defense.
Context
1.049M
Max output
943.718K
Input tokens
$4 / 1M tokens
Output tokens
$20 / 1M tokens