Architecture
data/models.yaml. Every label is auditable
against the model's sources.
Specs
- Architecture
- unknown
- Total params
- not disclosed
- Active params
- not disclosed
- Context window
- 200K tokens
- Attention
- unknown
- Position encoding
- unknown
- Post-training
- rlhf, constitutional
- OSI-approved
- no
- Data released
- no
- Training code
- not released
Benchmarks
Each score carries the date it was published; we never infer or interpolate missing scores.
General reasoning
| MMLU-Pro | 69.6 | as of 2026-05-21 | source ↗ |
| GPQA-Diamond | 48.9 | as of 2026-05-21 | source ↗ |
Code
| LiveCodeBench | 27.9 | as of 2026-05-21 | source ↗ |
Available quantizations
None. The weights are not distributed, so there are no public quantizations.
Notable innovations
- · 200K context window across the family
- · Vision input on every Claude 3 variant
- · Near-perfect Needle-in-a-Haystack recall (>99%)
Lineage
Predecessor flagship Opus; the Claude 4 line returned to the Opus name after skipping a 3.5 Opus.
Derivatives