OpenRouter 模型 ↗ · 2026-10-16
reflection/beam-1.0-501b-a23b
Model release OpenRouter 模型 Source language: Chinese
OpenRouter 于 2026-10-16 收录该模型(收录时间不等于模型正式发布日期);上下文:1,048,576 tokens;输入价:$0.3000/百万 tokens;输出价:$1.0000/百万 tokens;定位(来源描述):Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic workflows.
Full text
Reflection: Beam 1.0 501B A23B
reflection/beam-1.0-501b-a23b
Model weights
Compare
Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic workflows. It uses a sparse mixture-of-experts architecture with 501B total and 23B active parameters per token, and a 1M-token context window.
Beam targets software engineering, multi-step tool use, and mathematical and scientific reasoning, with an emphasis on token efficiency. It takes text input and produces text output, supports configurable reasoning effort, function calling and structured outputs, and is released under Apache 2.0.
Modalities
In / out price
$0.25 / $1per 1M
Context
1.0M
Released
Oct 16, 2026
Knowledge cutoff
Jun 2026
Compare
Providers
Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick: Balanced (price + speed), Nitro (fastest), or Floor (cheapest).
Latency / throughput
$0.30$1.00$0.040.51s145 tok/s
99.56%
$0.40$1.20$0.040.38s167 tok/s
99.76%
$0.25$1.00$0.051.07s131 tok/s
100.00%
$0.40$1.20$0.040.94s130 tok/s
99.33%
$0.40$1.20$0.040.18s395 tok/s
100.00%
ProviderInput / Output /M
Input $0.30 / Output $1.00
145 tok/s99.56% uptime
Input $0.40 / Output $1.20
167 tok/s99.76% uptime
Input $0.25 / Output $1.00
131 tok/s100.00% uptime
Input $0.40 / Output $1.20
130 tok/s99.33% uptime
Input $0.40 / Output $1.20
395 tok/s100.00% uptime
Pricing
The average price customers actually pay for this model, next to the prices providers post. Caching and discounts mean the price actually paid is often well below the listed one.
Weighted average
Input
$0.0683/M tokens
Output
$1.076/M tokens
Beam 1.0 501B A23B — Price history
Price history
No effective price data for this selection
$0.0529$1.0095.0%31.5%
$0.0619$1.2093.9%4.9%
$0.0641$1.0093.0%33.3%
$0.0898$1.2086.2%30.1%
$0.1690$1.2064.2%0.1%
ProviderInput / Output /M
Token share
Effective in $0.0529 / Effective out $1.00
Token share
31.5%
Effective in $0.0619 / Effective out $1.20
Token share
4.9%
Effective in $0.0641 / Effective out $1.00
Token share
33.3%
Effective in $0.0898 / Effective out $1.20
Token share
30.1%
Effective in $0.1690 / Effective out $1.20
Token share
0.1%
Performance
Throughput is how fast the model writes (tokens per second — higher is better). Latency is total round-trip time (lower is better). TTFT is time-to-first-token — how long before you see anything appear (lower is better).
Throughput
395tok/s
P50, best across providers
Latency
0.18s
P50, best provider
Reasoning effortLatency / throughput
AutoExacto benchmarks
GPQA DiamondTAU-BenchNebius Token Factory90.0%--DigitalOcean89.4%--CoreWeave89.3%--Baseten89.3%--Together88.6%--
+2 more providers
Uptime
Uptime is the percentage of the past 3 days that at least one provider was responding to requests. Availability is the percentage of time that inference was successfully served. OpenRouter continuously monitors and uses the next-best provider when one returns an error.
Uptime (3d)The model was reachable. Request routed to a provider.
100.00%
Availability (3d)The model returned inference from any provider. Errors and empty responses count against it.
99.90%
Availability over the last 3 days
Last 72 hours
Availability 99.90%
3 Days Ago2 Days AgoYesterdayNow
Availability over the last 24 hours
OpenRouter Availability
99.91%
Without Routing
99.41%
When an error occurs in an upstream provider, we can recover by routing to another healthy provider, if your request filters allow it. You can access per-provider uptime data programmatically through the Endpoints API. Learn more about our load balancing and provider routing options.
Benchmarks
Scores on standardized evaluations and this model’s own outputs on shared benchmark prompts. Higher percentages are better — and each rank shows where this model lands among the models evaluated on that source.
Benchmark score summary for Reflection: Beam 1.0 501B A23B (Artificial Analysis, Design Arena and Vals AI)
SourceBenchmarkScore
Artificial AnalysisBeam 1.0 Intelligence Index17.7
Artificial AnalysisBeam 1.0 HLE34.7%
Artificial AnalysisBeam 1.0 AA-LCR78.0%
Artificial AnalysisBeam 1.0 GDPval-AA33.9%
Artificial AnalysisBeam 1.0 CritPt13.1%
Artificial AnalysisBeam 1.0 SciCode48.5%
Artificial AnalysisBeam 1.0 Terminal-Bench 4.03.0%
Artificial AnalysisBeam 1.0 AA-Omniscience Accuracy30.7%
Artificial AnalysisBeam 1.0 AA-Omniscience Non-Hallucination Rate59.0%
Design ArenaBeam 1.0 (max) Models Arena 3D Elo1125
Design ArenaBeam 1.0 (max) Models Arena Code Categories Elo1224
Design ArenaBeam 1.0 (max) Models Arena Game Development Elo1187
Design ArenaBeam 1.0 (max) Models Arena UI Component Elo1260
Design ArenaBeam 1.0 (max) Models Arena Website Elo1239
Vals AIBeam 1.0 BioMysteryBench42.6%
Vals AIBeam 1.0 Code Migration v1.122.8%
Vals AIBeam 1.0 EMB v1.164.7%
Vals AIBeam 1.0 Finance Agent v2.151.8%
Vals AIBeam 1.0 Vals Finance Index46.5%
Vals AIBeam 1.0 HLAB-Vals-1.14.6%
Vals AIBeam 1.0 IOI22.8%
Vals AIBeam 1.0 Vals Legal Index61.8%
Vals AIBeam 1.0 Legal Research Bench28.8%
Vals AIBeam 1.0 MysteryMechanism22.1%
Vals AIBeam 1.0 ProofBench v1.118.0%
Vals AIBeam 1.0 Vals Science Index22.0%
Vals AIBeam 1.0 Tax Agent Bench8.8%
Vals AIBeam 1.0 Terminal-Bench 4.04.0%
Vals AIBeam 1.0 Terminal-Bench-Science1.4%
Vals AIBeam 1.0 Vals Index43.4%
Vals AIBeam 1.0 Vibe Code Bench v1.166.4%
Apps
Public apps that send the most traffic to this model. Good signal for what real production workloads look like — and a hint at which use cases this model is best suited for.
1.
Codex
A coding agent that helps you build and ship with AI
10.2Btokens
2.
Claude Code
Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, runs tests, and iterates on failures, all from natural language prompts.
5.18Btokens
3.
Hermes Agent
Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.
4.53Btokens
4.
Portkey AI
Control panel for AI apps
2.34Btokens
5.
pi
There are many coding agents, but this one is yours.
852Mtokens
Rest of OpenRouter
9.11B
Top 5 Apps
3.18B
Activity
Token volume and request traffic to this model over time.
API
Drop-in code to call this model. OpenRouter's API is OpenAI-compatible — most SDKs work by just swapping the base URL. The only thing that changes between models is the model slug below.
Not currently available through EU in-region routing. In-region routing docs
Frequently asked questions
Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic workflows. It uses a sparse mixture-of-experts architecture with 501B total and 23B active parameters per token, and a 1M-token context window. Beam targets software engineering, multi-step tool use, and mathematical and scientific reasoning, with an emphasis on token efficiency.
Beam 1.0 501B A23B costs $0.25/M input tokens and $1.00/M output tokens, with separate rates for Cache Read at $0.05/M tokens.
Beam 1.0 501B A23B has a 1,048,576 token context window. It supports up to 131,072 completion tokens.
Yes. Beam 1.0 501B A23B accepts tools and tool_choice for function calling on 4 of the 5 providers serving it, and requests that send tools are routed to those providers. It also supports structured outputs via a JSON schema in response_format.
Beam 1.0 501B A23B is served by 5 providers on OpenRouter: Nebius Token Factory, CoreWeave, DigitalOcean, Baseten, and Together. Requests are routed to the best available provider, with automatic failover to the others, and you can pin or exclude providers with provider routing.
Beam 1.0 501B A23B was released on October 16, 2026. Its knowledge cutoff is June 30, 2026.