Beam 1.0

Fiche modèle · Reflection AI · beam-1.0-501b-a23b

Beam 1.0 is Reflection AI's first open-weight model: a sparse Mixture-of-Experts reasoning model with 501.3B total parameters and 23.4B active per token (52 layers: 3 dense + 49 MoE; 256 experts; gated GQA with sliding-window attention; 152K vocabulary), pretrained on 23.8T tokens and post-trained with large-scale asynchronous RL. It targets coding, long-horizon agentic work, and tool use, with a 1M-token context window, up to 128K output tokens on the Reflection API, text-only input/output, and mandatory reasoning with low/medium/high/xhigh/max effort levels. Released under Apache 2.0 in BF16, FP8, and NVFP4 with a DFlash speculative-decoding drafter; available on Hugging Face, the Reflection API (Beam-1.0-501B-A23B), OpenRouter, and other partners.

Identité

OrganisationReflection AI
Pays—
Famille—
Typellm
Date de sortie2026-10-16
Licenceapache 2.0
Licence commercialeoui
Licence open sourceoui
Poids ouvertsoui
Auto-hébergement (DGX Spark)non — 501 Md > 107 Md
Modalitéstext
Fenêtre de contexte1 000 000
KV cache / token (Q8)106.5 ko/token
Paramètres totaux501 B
Paramètres actifs (session)23.4 B
Source des paramètres actifsapi : https://llm-stats.com/models/beam-1.0-501b-a23b
Curation des paramètres actifs2026-10-27
Prix par session0.64 $ (1 000 000 tokens entrée + 200 000 sortie, prix le moins cher par fournisseur)
Tokens / tâche (Data Players)1.2 M
Tokens / tâche (Terminal-Bench)83.3 M (est.)
Ratio Terminal-Bench / Data Players69.4×
Coût / tâche (Terminal-Bench)5.38 $ (est.)
Tokens d'entraînement23 800 000 000 000
Coupure de connaissance2026-06-30
Créé le2026-10-21T20:05:56.832266Z
Mis à jour le2026-11-04T06:05:12.317485Z
SourceLLM Stats
URLhttps://llm-stats.com/models/beam-1.0-501b-a23b

Prix par fournisseur

Statut du modèle (offres + auto-hébergement) : non — hors DGX Spark (> 107 Md).

Toutes sources confondues : catalogue llm-stats, routeurs EUrouter/Cortecs et catalogues directs (OVHcloud, Scaleway, Infomaniak). Les prix sont convertis en USD pour la comparaison ; le prix publié et le taux BCE utilisé sont dans l'infobulle. Les prix inconnus restent indiqués « — ».

FournisseurSourceSouverainetéEntrée (/M tok)Sortie (/M tok)QuantificationContexteLatence (s)Débit (tok/s)Statut
Reflection AIllm-statsinconnu——————active
Together AIllm-statsnon0.40 $1.20 $————active

Souveraineté qualifiée par offre (modèle × fournisseur) ; survolez le badge pour les critères.

Souveraineté

Scores par catégorie (top_scores)

CatégorieScore
instruction_following0.80
knowledge10.80
long_context0.70
math0.50
physics0.70
reasoning1.20
science10.80
search0.80
structured_output0.70
frontend_development0.80
general1.10
agents0.60
biology0.70
chemistry0.70
code0.70
tool_calling0.60
vision0.40

Scores par benchmark

BenchmarkCatégorieScoreMaxNormalisé (API)VérifiéAuto-déclaréDaté du
aa-lcrreasoning0.781.000.78nonoui2026-11-04T06:05:12.374647Z
aa-omniscience-indexknowledge10.80200.000.05nonoui2026-11-04T06:05:12.384025Z
aime-2026math0.981.000.98nonoui2026-11-04T06:05:12.338510Z
automationbenchagents0.391.000.39nonoui2026-11-04T06:05:12.359559Z
browsecompreasoning0.771.000.77nonoui2026-11-04T06:05:12.368396Z
critptreasoning0.161.000.16nonoui2026-11-04T06:05:12.350407Z
deepsearchqasearch0.811.000.81nonoui2026-11-04T06:05:12.371620Z
deepswe-1.1code0.431.000.43nonoui2026-11-04T06:05:12.320697Z
gpqareasoning0.911.000.91nonoui2026-11-04T06:05:12.353491Z
humanity's-last-examreasoning0.361.000.36nonoui2026-11-04T06:05:12.341725Z
humanity's-last-exam-(with-tools)reasoning0.511.000.51nonoui2026-11-04T06:05:12.344561Z
ifbenchinstruction_following0.801.000.80nonoui2026-11-04T06:05:12.381062Z
livecodebenchreasoning0.881.000.88nonoui2026-11-04T06:05:12.356714Z
longbench-v2long_context0.681.000.68nonoui2026-11-04T06:05:12.377590Z
mcp-atlasreasoning0.811.000.81nonoui2026-11-04T06:05:12.362574Z
scicodereasoning0.491.000.49nonoui2026-11-04T06:05:12.347715Z
swe-atlas-codebase-qnacode0.321.000.32nonoui2026-11-04T06:05:12.329657Z
swe-bench-multilingualreasoning0.781.000.78nonoui2026-11-04T06:05:12.332707Z
swe-bench-proreasoning0.661.000.66nonoui2026-11-04T06:05:12.323755Z
swe-bench-verifiedreasoning0.821.000.82nonoui2026-11-04T06:05:12.335623Z
tau3-bankingagents0.361.000.36nonoui2026-11-04T06:05:12.365657Z
terminal-bench-2.1reasoning0.801.000.80nonoui2026-11-04T06:05:12.326853Z

Classements

CatégorieRangScoreNote conservatriceBenchmarks évaluésMéthodeDaté duProvenance
code460.7033.789trueskill2026-11-04T08:08:39.742311Zrankings
knowledge2310.8011.451trueskill2026-11-04T08:08:45.139626Zrankings
math470.5035.125trueskill2026-11-04T08:08:42.442591Zrankings
reasoning441.2045.0619trueskill2026-11-04T08:08:41.093494Zrankings
agents53—28.759trueskill2026-10-27model_history

Lecture : Note conservatrice = borne basse TrueSkill (niveau estimé − marge d’incertitude) ; Benchmarks évalués = nombre de benchmarks de la catégorie pris en compte. À performances égales, un modèle évalué sur moins de benchmarks obtient une note plus basse : la note ne se déduit pas d’un seul score. Provenance : rankings = top 50 de /rankings ; model_history = historique du modèle (/models/{id}/history), utilisé pour les rangs au-delà du top 50.

Données

Dates de curation (récupération depuis l'API) : models 2026-11-04 · benchmarks 2026-11-04 · scores 2026-11-04 · updates 2026-11-04 · account 2026-11-04 · model_history 2026-11-04 · rankings 2026-11-04