Gemma
Google's open models — Gemini's research, small enough to self-host.
Gemma is Google DeepMind's open-weight family, built from the same research as Gemini but published so you can download and run it. Sizes run from a few billion parameters up to about thirty, they read text and images, and the current line has a 256K context with a switchable thinking mode. The obvious starting point if you want a capable model on your own machine or on cheap hardware.
At a glance — Gemma 3 27B
- Cost
- $0.08 in · $0.45 out per 1M tokensBudget
- Memory
- 131K tokens — about 200 pages
- Reads
- Text · Images
- Can do
- Use tools (agents)
- Structured output
- Thinking mode (not supported)
- Good at
- Coding: better than 4% · Agentic work: better than 0% · Overall: better than 1%
- Reliability
- 98.99% uptime, last 24h
- Released
- Mar 12, 2025
Advanced details
- API model ID
google/gemma-3-27b-it- Max output
- 118K tokens
- Cached input
- $0.04 per 1M
- Knowledge cutoff
- 2024-08-31
- Tokenizer
- Gemini
- Coding index
- 10.1
- Agentic work index
- 0.1
- Overall index
- 4.9
Providers
The same model, hosted by different companies. OpenRouter picks one per request and falls back to the next if it fails.
| Provider | Input /1M | Output /1M | Cached /1M | Max output | Uptime 24h |
|---|---|---|---|---|---|
| DeepInfra · fp8 | $0.08 | $0.16 | — | 16K | 98.84% |
| Parasail · fp8 | $0.08 | $0.45 | $0.04 | 118K | 98.99% |
| Nebius · fp8Degraded | $0.10 | $0.30 | — | 99K | 83.24% |
| Novita · bf16 | $0.119 | $0.20 | — | 16K | 92.78% |
Supported parameters
- frequency_penalty
- logit_bias
- logprobs
- max_tokens
- min_p
- presence_penalty
- repetition_penalty
- response_format
- seed
- stop
- structured_outputs
- temperature
- tool_choice
- tools
- top_k
- top_logprobs
- top_p
Live from OpenRouter, refreshed hourly. Scores from Artificial Analysis.
Key info
- Pricing
- Open source
- Category
- Models
- Best for
- Intermediate
- Made by
- Google DeepMind
- Weights
- Open
- Good for
- Running a capable model on your own machine