A newer version of this model is available: reflection/Beam-1.0-501B-A23B

MISSINGNO., the purple mystery jelly in a fake moustache, at a Nous Portal checkpoint.

Our token counts matched MISSINGNO. to the tokenizer family behind Beam on all 24 phrases, five days before Reflection named it.

Independent benchmark card by Stealth Models: our test results and findings for this model.

StealthMark results for the released Beam 1.0:

StealthMark 1.8.8 results for the released Beam 1.0: benchmark score 31.9 out of 100, General Intelligence 30.9, Visual Intelligence 33.8. Beam, the model behind MISSINGNO., is strongest at composing a whole scene: 52.4 on Beach.

Every StealthMark result of the released Beam 1.0: Beach 52.4, Foresight 49.4, Raccoon 46.3, Selfie 45.4, Latch 37.3, Ground Truth 27.9, Newton's Mirror 21.7, Mona Lisa 0.7.

Open each result · Visual Intelligence: Beach 52.4 · Raccoon 46.3 · Selfie 45.4 · Mona Lisa 0.7
General Intelligence: Foresight 49.4 · Latch 37.3 · Ground Truth 27.9 · Newton's Mirror 21.7

All results and comparisons on stealthmodels.com ↗

The case file. First seen 10 Oct 2026, revealed 16 Oct 2026, developer Reflection AI, context 262,144 tokens, input Text, tool calls Yes, including parallel calls, speed in our tests about 846 tok/s. 11 Oct 2026, our finding: Our token-count test matches all 24 phrases to the tokenizer family Beam turned out to use. 15 Oct 2026, our finding: The Cat Jam preview matches the same 24 counts in our test. 16 Oct 2026: Reflection releases Beam 1.0 with open weights and credits the MissingNo demo to it.

Sources: Reflection's Beam launch post · our token-count test

Facts as text
Fact Value
First seen 10 Oct 2026
Revealed 16 Oct 2026
Developer Reflection AI
Context 262,144 tokens
Input Text
Tool calls Yes, including parallel calls
Speed in our tests about 846 tok/s
Access Preview ended; the released Beam is available

Run it

Official weights of the released Beam reflection/Beam-1.0-501B-A23B · license: Apache 2.0, read it · GGUF: Unsloth · Local: 501B parameters, 23B active; this needs a large multi-GPU server.


Model guide · StealthMark · Play the benchmarks · Follow for new reveals
This card and its images: CC BY 4.0, credit stealthmodels.com. The model's weights have their own license. Updated 28 October 2026.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Space using stealth-model/missingno 1