Skip to navigation

Gemma 4 26B-A4B

Gemma 4 26B-A4B
Google · Text Generation
POST /v1/chat/completions

Gemma 4 26B A4B is a Google open multimodal model with 256K context, text, image, and video input, tools, and structured output.

See Gemma 4 26B-A4B pricing and specs or try it in the Playground.

At a glance

FieldValue
Model idgemma-4-26b-a4b
Model release date2026-03-31
Input modalitiesText, Image, Video
Output modalitiesText
Context window256K
Weight precisionFP8 weights + FP8 KV
Max output tokens32,768
Featuresreasoning, vision, video, function_calling, cache, multimodal, json_mode, logprobs
Native inferenceYes
NewNo
Structured outputJSON Schema
Supported endpointsPOST /v1/chat/completions, POST /v1/responses, POST /v1/messages, POST /v1/completions, POST /v1beta/models/gemma-4-26b-a4b:generateContent
Alternate model idsgemma-4-26b-a4b-it, google/gemma-4-26b-a4b, google/gemma-4-26b-a4b-it

Pricing

ChargeSpecRate
Inputper 1M prompt tokens$0.15 $0.05
Outputper 1M generated tokens$0.50 $0.29
Implicit cache readper 1M cached input tokens$0.15 $0.025
Web Search (Linkup)per call when invoked$0.013

Example request

curl https://api.empiriolabs.ai/v1/chat/completions \
-H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"model": "gemma-4-26b-a4b", "messages": [{"role":"user","content":"Hello"}]}'

Parameters

ParameterTypeRequiredDefaultDescription
temperaturenumberno1.0Sampling temperature. 0 is deterministic and 2 is maximum randomness. · Range: 0 – 2
top_pnumberno0.95Nucleus sampling probability mass. Lower values make outputs more focused. · Range: 0.01 – 1
max_tokensintegerno8192Maximum output tokens. When omitted: 8,192 at low and medium reasoning effort, 9,216 at high, and 32,768 at max or with thinking off. · Range: 1 – 32768
stopstringno-One or more stop strings.
reasoning_effortenumno"medium"Reasoning effort level. none disables thinking. low, medium, high, and max set bounded thinking budgets sized to the selected model. · Allowed: none, low, medium, high, max
thinking_budgetintegerno4096Maximum tokens reserved for reasoning when thinking is enabled. When max_tokens is lower, room is kept for the answer; a very low max_tokens can be spent entirely on reasoning. · Range: 128 – 32768
enable_thinkingbooleannotrueEnable reasoning before answering. With response_format, the model answers without reasoning unless the request sets enable_thinking.
top_kintegerno20Limit sampling to the top K candidate tokens when supported. · Min: 1
presence_penaltynumberno0Penalty for tokens that already appeared in the generated text. · Range: -2 – 2
frequency_penaltynumberno0Penalty based on how often a token has already appeared. · Range: -2 – 2
repetition_penaltynumberno1Higher values discourage repeated phrasing. · Min: 0.1
seedintegerno-Optional random seed for reproducible sampling. · Range: 0 – 2147483647
logprobsbooleannofalseReturn token log probabilities when supported.
top_logprobsintegerno-Return up to this many top token log probabilities. · Range: 0 – 20
logit_biasobjectno-Bias token IDs by adding positive or negative values before sampling.
toolsarrayno-OpenAI-compatible function tool definitions.
tool_choiceobjectno-OpenAI-compatible function tool selection.
response_formatenumno-Constrain the output to valid JSON. Use JSON mode for any JSON object, or supply a JSON Schema to force an exact response shape.
web_search_linkupbooleannofalseOptional web search powered by Linkup. When enabled, recent web sources are retrieved using your latest user message as the query and provided to the model as additional context. Adds $0.013 per call when invoked on top of the model’s normal token cost. Disabled by default.
disable_formattingbooleannofalseReturn the answer without the Sources list that web search adds.

Notes

Supports text, image, and video input, streaming, function tools, structured JSON output, seed control, and thinking mode on by default. Use reasoning_effort or thinking_budget for bounded thinking, or enable_thinking=false for direct answers. Automatic cache reads are billed at the cached-input rate when reported by the model service. Explicit cache controls are not supported.


Machine-readable schema: GET https://api.empiriolabs.ai/v1/models/gemma-4-26b-a4b.