Google: Gemma 4 26B A4B
google/gemma-4-26b-a4b-it
Model weightsGemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at a fraction of the compute cost. Supports multimodal input including text, images, and video (up to 60s at 1fps). Features a 256K token context window, native function calling, configurable thinking/reasoning mode, and structured output support. Released under Apache 2.0.
Providers
Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick: Balanced (price + speed), Nitro (fastest), or Floor (cheapest).
| $0.090 | $0.28 | $0.049 | 0.56s | 91 tok/s | ||
| $0.070 | $0.34 | -- | 0.36s | 61 tok/s | ||
| $0.090 | $0.30 | $0.050 | 0.48s | 70 tok/s | ||
| $0.130 | $0.40 | $0.050 | 0.44s | 88 tok/s | ||
| $0.067 | $0.29 | $0.039 | 0.74s | 68 tok/s | ||
| $0.140 | $0.40 | $0.050 | 1.07s | 58 tok/s | ||
| $0.100 | $0.30 | $0.050 | 0.52s | 45 tok/s | ||
| $0.050 | $0.15 | $0.025 | 0.65s | 39 tok/s | ||
| $0.100 | $0.30 | $0.050 | 0.25s | 25 tok/s | ||
| $0.040 | $0.15 | $0.010 | 0.61s | 46 tok/s | ||
| $0.070 | $0.24 | $0.046 | 0.95s | 75 tok/s | ||
| $0.060 | $0.33 | $0.040 | 0.57s | 72 tok/s | ||
| $0.130 | $0.40 | $0.050 | 0.87s | 66 tok/s | ||
| $0.150 | $0.60 | -- | 2.23s | 32 tok/s | ||
| $0.130 | $0.40 | -- | 0.62s | 109 tok/s |
- Input $0.09 / Output $0.2891 tok/s100.00% uptime
Input $0.07 / Output $0.3461 tok/s100.00% uptime- Input $0.09 / Output $0.3070 tok/s100.00% uptime
- Input $0.13 / Output $0.4088 tok/s100.00% uptime
- Input $0.067 / Output $0.2968 tok/s99.91% uptime
- Input $0.14 / Output $0.4058 tok/s99.91% uptime
- Input $0.10 / Output $0.3045 tok/s100.00% uptime
- Input $0.05 / Output $0.1539 tok/s100.00% uptime
- Input $0.10 / Output $0.3025 tok/s97.03% uptime
Input $0.04 / Output $0.1546 tok/s99.97% uptime- Input $0.07 / Output $0.2475 tok/s100.00% uptime
- Input $0.06 / Output $0.3372 tok/s99.88% uptime
- Input $0.13 / Output $0.4066 tok/s100.00% uptime
- Input $0.15 / Output $0.6032 tok/s91.36% uptime
- Input $0.13 / Output $0.40109 tok/s84.61% uptime
Pricing
The average price customers actually pay for this model, next to the prices providers post. Caching and discounts mean the price actually paid is often well below the listed one.
Weighted average
Gemma 4 26B A4B — Price history
| $0.0363 | $0.120 | 16.8% | 8.4% | ||
| $0.0372 | $0.147 | 8.4% | 5.5% | ||
| $0.0470 | $0.327 | 64.5% | 4.1% | ||
| $0.0614 | $0.288 | 19.7% | 4.3% | ||
| $0.0626 | $0.237 | 29.7% | 1.3% | ||
| $0.0683 | $0.255 | 26.7% | 26.8% | ||
| $0.0698 | $0.337 | 0.0% | 4.2% | ||
| $0.0821 | $0.278 | 18.9% | 30.0% | ||
| $0.0869 | $0.298 | 25.6% | 2.8% | ||
| $0.0965 | $0.299 | 6.9% | 4.6% | ||
| $0.1083 | $0.398 | 26.7% | 4.0% | ||
| $0.1227 | $0.398 | 9.0% | 0.6% | ||
| $0.1260 | $0.399 | 15.4% | 0.3% | ||
| $0.1299 | $0.400 | 0.0% | 1.8% | ||
| $0.1499 | $0.599 | 49.9% | 1.3% |
- Effective in $0.0363 / Effective out $0.120Token share8.4%
- Effective in $0.0372 / Effective out $0.147Token share5.5%
- Effective in $0.0470 / Effective out $0.327Token share4.1%
- Effective in $0.0614 / Effective out $0.288Token share4.3%
- Effective in $0.0626 / Effective out $0.237Token share1.3%
- Effective in $0.0683 / Effective out $0.255Token share26.8%
- Effective in $0.0698 / Effective out $0.337Token share4.2%
- Effective in $0.0821 / Effective out $0.278Token share30.0%
- Effective in $0.0869 / Effective out $0.298Token share2.8%
- Effective in $0.0965 / Effective out $0.299Token share4.6%
- Effective in $0.1083 / Effective out $0.398Token share4.0%
- Effective in $0.1227 / Effective out $0.398Token share0.6%
- Effective in $0.1260 / Effective out $0.399Token share0.3%
- Effective in $0.1299 / Effective out $0.400Token share1.8%
- Effective in $0.1499 / Effective out $0.599Token share1.3%
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).
P50, best across providers
P50, best provider
Uptime
Uptime is the percentage of the past 3 days that at least one provider was responding to requests. Availability is the percentage of time that inference was successfully served. OpenRouter continuously monitors and uses the next-best provider when one returns an error.
Availability over the last 3 days
Last 72 hoursAvailability over the last 24 hours
When an error occurs in an upstream provider, we can recover by routing to another healthy provider, if your request filters allow it. You can access per-provider uptime data programmatically through the Endpoints API. Learn more about our load balancing and provider routing options.
Benchmarks
Scores on standardized evaluations and this model’s own outputs on shared benchmark prompts. Higher percentages are better — and each rank shows where this model lands among the models evaluated on that source.
| Source | Benchmark | Score |
|---|---|---|
| Artificial Analysis | Gemma 4 26B A4B (Non-reasoning) GPQA Diamond | 71.4% |
| Artificial Analysis | Gemma 4 26B A4B (Non-reasoning) HLE | 11.5% |
| Artificial Analysis | Gemma 4 26B A4B (Non-reasoning) IFBench | 45.4% |
| Artificial Analysis | Gemma 4 26B A4B (Non-reasoning) τ²-Bench Telecom | 40.4% |
| Artificial Analysis | Gemma 4 26B A4B (Non-reasoning) AA-LCR | 42.3% |
| Artificial Analysis | Gemma 4 26B A4B (Non-reasoning) CritPt | 0.0% |
| Artificial Analysis | Gemma 4 26B A4B (Non-reasoning) Terminal-Bench Hard | 25.0% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) Coding Index | 39.3 |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) Agentic Index | 2.9 |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) GPQA Diamond | 79.2% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) HLE | 19.3% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) IFBench | 72.4% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) τ²-Bench Telecom | 43.6% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) AA-LCR | 65.7% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) τ-Bench Banking | 12.0% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) GDPval-AA | 3.6% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) CritPt | 0.0% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) SciCode | 40.0% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) Terminal-Bench Hard | 13.6% |
| Artificial Analysis | Gemma 4 26B A4B (Reasoning) Terminal-Bench 2.1 | 39.0% |
Apps
Public apps that send the most traffic to this model. Good signal for what real production workloads look like — and a hint at which use cases this model is best suited for.
Activity
Token volume and request traffic to this model over time.
API
Drop-in code to call this model. OpenRouter's API is OpenAI-compatible — most SDKs work by just swapping the base URL. The only thing that changes between models is the model slug below.
Not currently available through EU in-region routing. In-region routing docs
Frequently asked questions
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at a fraction of the compute cost. Supports multimodal input including text, images, and video (up to 60s at 1fps).
Gemma 4 26B A4B costs $0.04/M input tokens and $0.15/M output tokens, with separate rates for Cache Read at $0.01/M tokens.
Gemma 4 26B A4B has a 262,144 token context window. It supports up to 131,072 completion tokens.
Yes. Gemma 4 26B A4B accepts tools and tool_choice for function calling on 12 of the 15 providers serving it, and requests that send tools are routed to those providers. It also supports structured outputs via a JSON schema in response_format.
Gemma 4 26B A4B accepts images, text, and video as input and returns text.
Gemma 4 26B A4B is served by 15 providers on OpenRouter: Darkbloom, Prime Intellect, DekaLLM, io.net, AkashML, DeepInfra, Makora, NextBit and 7 more. Requests are routed to the best available provider, with automatic failover to the others, and you can pin or exclude providers with provider routing.
Gemma 4 26B A4B was released on April 3, 2026.