Motif 3 by Motif Technologies is a 314B sparse MoE with 13B active, 256K native context, Grouped Differential Latent Attention, and self-speculative decoding.
Capabilities, design details, and architectural traits
Motif 3 is a large-scale Mixture-of-Experts language model built from the ground up by Motif Technologies, explicitly described as a fully in-house proprietary design rather than a re-parameterization of existing open-source architectures. The model introduces several custom components not found in other major MoE systems.
| Trait | Detail |
|---|---|
| Grouped Differential Latent Attention (GDLA) | Custom attention mechanism designed in-house for this model, not derived from existing architectures |
| Grouped PolyNorm activation | Novel activation function applied per expert, specific to Motif 3's architecture |
| Multi-Token Prediction (MTP) head | Single-layer MTP head built into the model, enabling self-speculative decoding during inference |
| 384 routed experts, top-8 plus 1 shared | Sparse routing activates 8 of 384 experts per token with an additional shared expert, yielding ~13B active of ~314B total parameters |
| 256K native context | 262,144-token context window built natively into the architecture |
| Modified mHC | Custom modification of mHC component, part of the in-house design |
The built-in MTP head allows Motif 3 to perform self-speculative decoding without a separate draft model. In vLLM, this is activated through a --speculative-config flag with num_speculative_tokens: 1 identified as optimal for this model. The model also ships with custom reasoning and tool-call parsers (--reasoning-parser motif, --tool-call-parser motif) for structured output.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Motif Technologies | Anthropic | Anthropic |
| Release Date | July 14, 2026 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | - | May 2026 | - |
| Context & Limits | |||
| Context Window | 262K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $5 | $10 |
| Output Pricing | Free Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | text | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 45.3 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | 62.0 | 78.0 Best Coding Index | 76.5 |
| Agentic Index | 34.9 | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
