Reflection Beam
October 16 open-weight text reasoning and coding model: 501.3B total / 23.4B active parameters, 1M context and BF16/FP8/NVFP4 checkpoints.
- Practical guide
Access choices, version boundaries and evaluation limits.
- Official model source
Verify the exact checkpoint or provider contract.
- Official model source
Verify the exact checkpoint or provider contract.
- Official model source
Verify the exact checkpoint or provider contract.
Model Specificationsreflection
Large-memory GPU inference; BF16, FP8 and NVFP4 are separate checkpoints.
Self-hosting infrastructure costs vary. Confirm any hosted rate directly with the provider.
Scoped Terms🚀Function Overview
Primary-source contract checked October 21, 2026. No independent model request or hardware test was executed.
Key Features
- Text input and text output; predominantly English.
- Apache 2.0 checkpoints; numerical format and inference implementation must match.
- 1M-token model context; no universal provider output cap or token price is asserted.
- Hosted access and self-hosted weights have separate runtime and service contracts.
Use Cases
- •Compare the exact checkpoint and provider contract
- •Evaluate a permitted task with a verifiable reference
⚙️Input Parameters
model
stringreflection/Beam-1.0-501B-A23B — follow the selected official runtime or provider interface; this is not a universal hosted alias.
task_input
provider-specificUse the documented input schema; see official sources for supported modalities and request fields.
Quick Actions
Technical Specifications
- Hardware Type
- Large-memory GPU inference; BF16, FP8 and NVFP4 are separate checkpoints.
- Commercial Use
- Apache 2.0 for the published checkpoint; hosted service terms remain separate.
- Pricing
- Self-hosting infrastructure costs vary. Confirm any hosted rate directly with the provider.