Open Model WeightsREFLECTION · FIELD VERIFIED
Primary sources last checked on 18 Oct 2026. Verification freshness is tracked separately from factual change, and unknown values remain explicit rather than inferred.
FRESHNESS & VERIFICATION
A verification check does not pretend the model changed. Open Model Weights updates the verification date after a real primary-source check, while factual-update dates move only when retained source evidence changes.
CONFLICT INTELLIGENCE
+ Save to my shortlistMonitor changes →Local browser monitoring · no background email alerts
DIRECT ANSWER · FEATURED MODEL BRIEFING
Source-grounded editorial layer · reviewed 2026-10-18
Beam DFlash2 companion drafter is a publisher-released open-weight checkpoint in the Beam DFlash2 family. DFlash2 is a small companion draft model for speculative decoding with the much larger Beam verifier. It is not a standalone 501B-reasoning chatbot or a replacement for the Beam weights. OMW verified its exact publisher repository revision, 1 recognized weight artifacts, apache-2.0 license evidence and 1,048,576 configured tokens on 18 October 2026. This source-based review does not represent an independent model inference, speed or quality benchmark.
WHY IT MATTERS
Beam separates a substantial sparse MoE model from the BF16, FP8 and NVFP4 formats used to deploy it. The 23.4B-active claim should never be misread as a 23.4B total weight file.
Apache-2.0 weights and a published speculative draft component create useful deployment choices, but do not establish independent inference speed or benchmark superiority.
MODEL-SELECTION VIEW
The DFlash2 repository includes the companion drafter rather than a substitute for the main Beam model. A valid speculative generation stack must run the large Beam verifier to accept or reject draft tokens; quoting 1.93B drafter parameters as the footprint of the combined system would be false.
Evaluate acceptance rate, total memory for both models and end-to-end latency against Beam-only decoding on the same prompts. Speculative speedup depends on implementation and workloads, and no OMW benchmark has been executed.
The publisher discusses Beam for coding, reasoning and agent tasks. Its published config supplies 1,048,576 context tokens; these values are source-verified declarations, not an OMW benchmark of very long agent sessions.
A controlled test should measure prompt prefill, cache allocation and output quality as requests grow. Actual inference length also depends on kernel support, context parallelism, generation length and the presence of concurrent requests.
Reflection publishes separate repositories for high-precision BF16, FP8 and NVFP4 checkpoints. They are not merely URL aliases: each has a distinct set of tensor artifacts, revision SHA and potential serving dependencies.
Moving to a lower precision may reduce stored weight bytes, but neither an unchanged model name nor a Safetensors index is evidence of equal numerical results or higher real-world throughput. Compare task outputs and hardware memory under the same prompts and decoding policy.
The verified source repositories include Apache-2.0 licensing evidence. Commercial and derivative use is generally allowed under the license's terms, including applicable notices and patent provisions; verify the exact file and redistributed weights.
Model documentation and training-related source material are separate from actual inference execution. OMW links checked model cards, versions and licensing while making no claim that it has trained, fine-tuned or served Beam.
Although the DFlash2 record includes a config context field, speculative decoding requires the larger verifier to validate draft tokens. Do not market this companion checkpoint as delivering the full 501B-capability reasoning experience or the same runtime behavior.
Its value lies in the accepted-token rate and the total deployment cost alongside the full model. Document the verifier revision, speculative settings and any failure to preserve its output distribution.
IMPLEMENTATION & EVALUATION
Reproduction checklist for Beam DFlash2 companion drafter, not an OMW inference run. Keep the exact BF16 checkpoint, tokenizer, media processing, tool harness, software revision and hardware in the report.
Record the official repository SHA, artifact manifest and applicable weight-specific license before benchmarking.
Save a reproducible revision and artifact checksum list together with the exact license source and date.
Model branches may change, and third-party conversions or processors can alter the tested input without notice.
Reproduce the model-card tokenizer, multimodal preprocessing, serving code, precision and tool-call format.
Run at least one positive and one failure-mode task with pinned environment and exact model implementation, or record incompatibility.
Declared library tags do not prove a given package version supports this architecture or compressed format.
Use representative prompt lengths, media dimensions and tool interactions while instrumenting actual memory, latency and correctness.
Retain raw input/output pairs, peak memory, prefill/decode latency, error logs and infrastructure specification.
A configured context limit, weight-only arithmetic or publisher chart is not an independent memory or throughput benchmark.
Run the same acceptance set on a source-linked alternative with aligned prompt budgets and evaluation rules.
Publish both revision IDs, model formats, test harness, pass/fail criteria, observed variance and unsupported cases.
Different engines, sampling strategies, modal input processing and agent permissions confound naive rankings.
Proposed tests, not OMW benchmark results. Read the official model card at the pinned source revision ↗ Inspect OMW field evidence ↗
PRIMARY & INDEPENDENT SOURCES
FAQ
No. The 1.93B companion is a speculative draft producer; the larger Beam model must verify proposed tokens. It is not a 501B reasoning checkpoint.
The original Reflection repositories include Apache-2.0 licensing evidence. Read the checked license and comply with its terms before redistributing or using derivatives.
The source configuration declares 1,048,576 tokens, but OMW has not measured full-length inference latency, memory or successful output quality.
They are distinct first-party checkpoints with different stored numerical representations, artifacts and potential runtime requirements. Performance must be benchmarked separately.
No. The sparse MoE routing figure is not the total parameter storage, nor a complete measurement of KV cache and runtime allocations.
No. Coding and agent claims in the publisher material are not reproduced OMW test results.
MODEL INTELLIGENCE · SOURCE-FIRST
Open Model Weights records Beam DFlash2 companion drafter as a 1.93B (publisher-declared)-parameter speculative draft for beam; not standalone open-weight model in the Beam DFlash2 family, with Reflection listed as the developer. The checked source repository provides 1 recognized weight artifacts, format evidence for Safetensors and a verified context window of 1,048,576 tokens. Values that are not evidenced in the checked sources remain explicitly undisclosed or unknown.
The repository declares apache-2.0. Commercial use allowed under apache-2.0 conditions. Open Model Weights reports the checked license evidence and classification; this is not legal advice.
The checked sources declare reflection/Beam-1.0-501B-A23B as lineage evidence. Training recipe status: not-disclosed-in-checked-repository; data disclosure status: not-disclosed-in-standard-metadata-or-obvious-model-card-section.
DEPLOYMENT PROFILE
These bars visualize weight-only decimal-GB estimates. They exclude KV cache, activations, runtime overhead, optimizer state and sharding.
LICENSE & USE
Commercial use allowed under apache-2.0 conditions
This section summarizes the checked repository evidence. It is not legal advice.
LINEAGE & PROVENANCE
Open Model Weights does not infer a hidden base model or training corpus when the checked sources do not declare one.
RUNTIME SIGNALS
Declared and model-card-mentioned runtimes are discovery signals. They are not treated as Open Model Weights runtime tests unless explicitly measured.
EVIDENCE TRAIL
This is where the article remains auditable: source links on one side, the registry's observed change history on the other.
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
01 · WEIGHTS
02 · LICENSE
Classification describes the checked license terms; it is not legal advice.
03 · HARDWARE
Weight-only decimal-GB estimate from total parameter count; excludes KV cache, activations, runtime overhead, optimizer state and sharding.
04 · FORMATS
“Not observed” means no matching repository artifact/name/dtype signal was found; it does not prove no third-party conversion exists.
05 · LINEAGE
06 · TRAINING ASSETS
07 · RUNTIME SUPPORT
Open Model Weights separates model-repository evidence from external runtime-release evidence. An external support statement does not change the official model repository's weight formats and does not imply an Open Model Weights inference benchmark.
08 · SOURCES
09 · VERIFICATION HISTORY
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.