Post-Cutoff

Reflection AIBeam familyCurrentReasoning modelReleased 5 Oct 2026

Beam (Beam-501B-A23B)

Reflection AI’s first model. Until the weights ship, access is through the waitlisted beta API. The OpenAI-compatible endpoint works with the official OpenAI SDKs once you change base_url to https://api.reflection.ai/openai/v1 and use a Reflection API key. When the weights are released, add the Hugging Face repo and switch status to current.

Model idBeam-501B-A23B

PricePrice not published

Checked17 October 2026

Knows up to June 2026

1,923events after its cutoff

84 historic, 451 major. 142 days of news.

It launched on 5 October 2026, already missing 686 of them.

Five things it has never heard of

How to call it

5 ways to use Beam (Beam-501B-A23B), with the model id each provider expects.

Providers, model ids, endpoints and docs for Beam (Beam-501B-A23B)
ProviderModel idEndpointDocs
Reflection API (beta, early access / waitlist) Beam-501B-A23B https://api.reflection.ai/v1 developers.reflection.ai
Reflection API, OpenAI-compatible (Chat Completions, Models) Beam-501B-A23B https://api.reflection.ai/openai/v1 developers.reflection.ai
Hugging Face (open weights, Apache 2.0; BF16, FP8, NVFP4 and DFlash2 variants) reflection/Beam-1.0-501B-A23B huggingface.co
OpenRouter (served by Baseten; price not listed at launch per VentureBeat)
Web app / early-access signup platform.reflection.ai

Pricing

Price not published.

No price source linked. Checked 17 October 2026.

Specs

Input
Text
Output
Text
Context window
256,000 tokens
Max output
128,000 tokens
Open weights
Yes
Licence
Apache-2.0
Training cutoff
June 2026
Released
5 October 2026
Status
Current
Type
Reasoning model

What stands out

What its maker marketed at launch, and what people found later, each with its source.

  • Efficient 500B-class open MoE

    501B total / 23B active sparse MoE trained on 23.8T tokens; Reflection says it matches GLM-5.2 on reasoning with 3-4x less inference compute.

    Source: reflection.ai

  • Large-scale agentic RL

    RL on 10.5K GB300 GPUs for 4 weeks: >100M rollouts across ~1M coding, agentic and STEM environments using ~1.3B sandboxes.

    Source: reflection.ai

  • 1M-token context (training)

    Pretrained at 256K and extended to 1M tokens in midtraining; the beta API currently exposes 256K context and 128K output.

    Source: developers.reflection.ai

Four assistants, one scale

Beam (Beam-501B-A23B) is added as a fifth row. Each row runs from what the model knows (graphite) through the months between its cutoff and its launch (hatched) to today (pink).

  1. GPT-6.2 Astra 1,915 at launch, 1,958 today
  2. Claude Opus 5.5 372 at launch, 1,923 today
  3. Gemini 4 Hydrogen 1,880 at launch, 2,018 today
  4. Grok 4.7 388 at launch, 1,958 today
  5. Beam (Beam-501B-A23B) 686 at launch, 1,923 today
Show as a table
ModelCutoffReleasedMissing at launchMissing today
GPT-6.2 AstraMay 202617 November 20261,9151,958
Claude Opus 5.5June 202622 September 20263721,923
Gemini 4 HydrogenMarch 202612 November 20261,8802,018
Grok 4.7May 202621 September 20263881,958
Beam (Beam-501B-A23B)June 20265 October 20266861,923

Notes

Announced Oct 5, 2026; Apache-2.0 weights released on Hugging Face Oct 16, 2026 (reflection/Beam-1.0-501B-A23B, plus FP8/NVFP4 quantizations; 8-bit and 4-bit versions fit one 8-GPU server per Reflection), with a final model card and the Beam Safety Technical Report. Reflection hosted API free during public beta. API is beta, reasoning always on (default effort medium), tool calling and structured outputs. No published price. Benchmarks are Reflection’s own: SWE-bench Verified 80.9, Terminal Bench v2.1 80.1, GPQA Diamond 90.5, HLE no tools 36.2.