比较服务商

每个点是一家服务商。从左到右是每百万 token 的输入价,从下到上是输出价。

基础价格

5
最新价格检查于 Nov 18, 2026。

Beam 1.0 501B A23B · 已公布价格

Hosted API prices · USD per 1M tokens
ProviderInputOutputCached inputScope
ReflectionNot announcedNot announcedNot announcedFirst-party public tariff not found
Nebius via OpenRouter$0.25$1.00Not announcedHosted-provider offer
Together via OpenRouter$0.40$1.20$0.04Hosted-provider offer
[2][3][4]

Beam 1.0 501B A23B · 权重许可证

权重许可证
模型权重协议
开放权重Apache 2.0
[1]

Beam 1.0 501B A23B · 权重与量化

Apache 2.0 deployment and fine-tuning
ItemConfiguration / evidence
WeightsBF16 / FP8 / NVFP4; official Hugging Face checkpoints
[1][5][6]

Beam 1.0 501B A23B · 部署方式

Apache 2.0 deployment and fine-tuning
ItemConfiguration / evidence
Self-hostingDownload weights → SGLang / vLLM / TokenSpeed → Beam Verifier
[1][5][6]

Beam 1.0 501B A23B · 微调支持

Apache 2.0 deployment and fine-tuning
ItemConfiguration / evidence
Fine-tuningLoRA / QLoRA / SFT / RL; stack-specific recipes coming soon
[1][5][6]

Beam 1.0 501B A23B · 已知风险与限制

Apache 2.0 deployment and fine-tuning
ItemConfiguration / evidence
EfficiencyReflection reports about 3× fewer reasoning tokens; no matched GLM/Kimi efficiency result verified
Evidence limitsVendor benchmarks use their own harnesses; long-horizon production stability is not established
[1][5][6]

Beam 1.0 501B A23B · 使用场景

Apache 2.0 deployment and fine-tuning
ItemConfiguration / evidence
Agent integrationMirror CLI; coding, terminal tasks and tool use
[1][5][6]

上下文与限制

  • Reflection Beam-1.0-501B-A23B has no first-party token tariff in the reviewed public documentation.

    [2][3][4]

关于 Beam 1.0 501B A23B

来源与注脚

  1. [1]
  2. [2]
    Reflection official launch核验日期:
  3. [3]
    Reflection API models核验日期:
  4. [4]
    OpenRouter hosted-provider offers核验日期:
  5. [5]
    Reflection official documentation核验日期:
  6. [6]
    Reflection official documentation核验日期:

输入、输出和缓存读取的单位是每百万 token 的美元价格。速度是每秒输出 token 数。可用率是来源报告的在线比例。