Gemma 4 26B MoE
#48 v Open-source jazykové modelygoogle · v4 · 26b moe · od 2026-04-02 · 2× · naposledy 30. 6. 2026
9
Momentum
Gemma 4 26B MoE is a Mixture-of-Experts language model with 26 billion parameters. According to the excerpts, it is positioned for high-throughput reasoning and represents the mid-tier performance level between edge models (2B/4B) and a dense 31B model.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Benchmark Score (MMLU/Similar) | MMLU Pro: 82.6% | AIME 2026: 88.3% | GPQA Diamond: 82.3% | LiveCodeBench v6: 77.1% | Arena-AI ELO (Text): 1441 |
| Context Window | 256,000 tokens (262,144 token context window) |
| Model Size (Parameters) | 25.2B total parameters (MoE); 3.8B active parameters per token at inference |
| Price Tier | Open-weight / free (self-hosting, Apache 2.0); API via OpenRouter: $0.06/1M input tokens, $0.33/1M output tokens (free tier also available) |
| Memory Requirement | ~15 GB VRAM at Q4 quantization (minimum 14 GB Q4 / 28 GB BF16); all 26B total parameters must be fully loaded into memory |