Gemma 4 26B A4B IT
google/gemma-4-26b-a4b-it- Family · Gemma 4
- Open weights
- Image · Text · Video → Text
- Tools · JSON · Vision
Not on OSMEXO yet
Spec & list price
Context256K
Max output236K
Input / 1M$0.08
Output / 1M$0.263.3× input
Usage
Tokens / day48.3B▼ 5.1% 7d
Usage rank#49
Requests / day14.7M
Tool calls / req0.02
Per request · 30 days
Avg input2.7K0.2× median
Avg output1840.3× median
Cache hit24%0.3× median
Reasoning share7.6%0.1× median
Tool-call failures1%0.5× median
Consumption
Peak 62.8B 8 OctLow 37.1B 1 NovRank today #49
News & trends
18 mentions this week ▼ 14%Gemma 4 family · videos, posts, articles · 30 days
Trending · 14 days · Gemma 4 family
- Gemma-4-31B-Humanlike-Chat: 2.0's improved human voice on official Gemma 4 31B319 upvotes
- Qwen 3.8 27B vs Gemma 4 31B on RTX 5090: 4 Real Coding Tests2.3K views
- Qwen 3.8 27B vs Gemma 4 31B on RTX 5090: 4 Real Coding Tests7.6K views
- Gemini 3.6 Flash won a nuclear last stand without breaking a single pact. Google's open Gemma 4 31B came 2nd of 43.69 upvotes
News · latest from Google
- Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v529 Oct
- Gemini 4 Carbon (High) Scores 59 on Artificial Analysis v528 Oct
- Gemini 4 Carbonblog.google · 28 Oct
- Gemini 4 Carbon: Our next frontier modelGoogle DeepMind · 28 Oct
Benchmarks
Benchmark data: Artificial AnalysisOutput speed#170/33466 tok/sabout median
Fastest provider149 tok/sbest of 16 providers
First token#920.63 s1.6× quicker than median
Family progress
Intelligence Index
- Family releases14.2
Technical details
- Tokenizer
- Gemma
- License
- Apache 2.0
- Parameters
- 25.8B
- Moderation
- none
- Downloads · likes
- 5.9M · 1.7K
Request parameters 20
- tools
- tool_choice
- response_format
- structured_outputs
- reasoning
- include_reasoning
- temperature
- top_p
- top_k
- min_p
+10 more
- max_tokens
- max_completion_tokens
- stop
- seed
- frequency_penalty
- logit_bias
- logprobs
- presence_penalty
- repetition_penalty
- top_logprobs
When to use
Best for
- Self-hosted chat with image and video input
- Cost-efficient inference with a mixture-of-experts design
- Private deployments with tool calling
- Fine-tuning a capable open model
Think twice for
- Very long documents beyond its context window
- Hardest frontier reasoning tasks
- Audio input
Research papers
- Are You Reading the Right Thing? Rethinking Internal Signals for Prompt-Injection Detection14 Oct
- From Routing Signals to Selective Review: Visual regrounding in MoE VLMs29 Sep
- Emergent Goal-Directed Attention in Large Vision-Language Models31 Aug
- Natural Language Code Retrieval for 1C:Enterprise: An Open Benchmark and Efficient Bi-Encoder20 Aug
- Abliteration Is Not a Scalpel: Off-Target Effects of Refusal Removal on Decision Disposition Across Model Families19 Jul
- Architecture-Sensitive Supervised Fine-Tuning for Screen-Conditioned Action Prediction: A PiSAR Benchmark28 May
- TACT: Mitigating Overthinking and Overacting in Coding Agents via Activation Steering7 May
Same family
Gemma 4- 8.9
- 14.7
- 14.2
- 16.7
- 7.8
Want Gemma 4 26B A4B IT on OSMEXO?
Sign in to request it. Requests are reviewed; a request does not guarantee availability.