DiffusionGemma 26B A4B IT
Released
2026-06-10
Last refreshed
2026-06-12
Status
Researched 134d ago
Open sourceCommercial use: permittedMultimodalCodingLong contextVisionClassification
DiffusionGemma 26B A4B IT is available now for coding, long context, and vision with open-source and 256K context; evaluate it while provider pricing coverage matures.
Google DeepMind releases · 53 in the last 12 months · this family litChangelog →
Specifications
- Family
- DiffusionGemma
- Released
- 2026-06-10
- Context
- 256K
- Parameters
- 26B
- Architecture
- Mixture of Experts
- Specialization
- general
- Openness
- Open source
- License
- Apache 2.0OSI-approvedPermitted
- Weights
- Available
- Code
- Unknown
- Training
- Fine-tuned
Created by
Pricing
No tracked provider pricing is available yet.
Links
About
Instruction-tuned DiffusionGemma model built on the Gemma 4 26B-A4B backbone. Uses a masked-diffusion (non-autoregressive) inference strategy for faster text generation — the model iteratively refines a sequence of masked tokens rather than predicting each token left-to-right. The 26B sparse MoE architecture activates approximately 4B parameters per forward pass. Released open-source under Apache 2.0 on June 10, 2026.
Providers
No active provider routes are tracked for this model yet.
Capabilities
VisionMultimodalReasoningFine-tuning
Benchmark peer barsfor Coding
LiveCodeBenchRank 55 of 75
Benchmark scores(8)
Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
| Benchmark | Score | Version | Evaluation | Source |
|---|---|---|---|---|
| 77.6 | MMLU ProObserved 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: Static eval Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'MMLU Pro'. Retrieved 2026-10-19. | ||
| 69.1 | AIME 2026, no toolsObserved 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: No tools Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'AIME 2026 no tools'. Retrieved 2026-10-19. | ||
| 69.1 | LiveCodeBench v6Observed 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: Static eval Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'LiveCodeBench v6'. Retrieved 2026-10-19. | ||
| 1429.0 | Codeforces ELOObserved 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: Static eval Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'Codeforces ELO'. Retrieved 2026-10-19. | ||
| 73.2 | diamondObserved 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: Static eval Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'GPQA Diamond'. Retrieved 2026-10-19. | ||
| 11.0 | HLE, no toolsObserved 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: No tools Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'HLE no tools'. Retrieved 2026-10-19. | ||
| 11.9 | HLE, with searchObserved 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: With search tool (card does not detail harness) Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'HLE with search'. Retrieved 2026-10-19. | ||
| 54.3 | MMMU ProObserved 2026-10-19 | Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler) Harness: Static eval Evaluator: Google DeepMind Confidence: self-reported Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'MMMU Pro'. Retrieved 2026-10-19. |
Migration checks
No linked migration route is available for this model yet.
Created by
Pricing
No tracked provider pricing is available yet.
Links