Beam

Coding

• SWE-bench Verified#2 of 18
  80.9%vendorReflection AI: Introducing Beam ↗

  Leader: Claude Sonnet 5 at 85.2%

• SWE-bench Pro#8 of 33
  65.5%vendorSWE-bench Pro v1Reflection AI: Introducing Beam ↗

  Leader: Claude Opus 5.5 at 89.9%

• Terminal-Bench 2.x#24 of 105
  80.1%vendorTerminal-Bench 2.1, vendor harnessReflection AI: Introducing Beam ↗

  Leader: Claude Fable 5.1 at 91.4%

• SciCode#48 of 100
  49.7%vendorReflection AI: Introducing Beam ↗

  Leader: Claude Opus 5.5 at 66.9%

Agents and tool use

• MCP Atlas#3 of 10
  78.7%vendorReflection AI: Introducing Beam ↗

  Leader: Kimi K3 at 84.2%

• BrowseComp#8 of 16
  77.4%vendorwith context managementReflection AI: Introducing Beam ↗

  Leader: GPT-6 Astra at 91.5%

Reasoning and knowledge

• AIME 2026#3 of 15
  97.8%vendorReflection AI: Introducing Beam ↗

  Leader: ERNIE 5.1 at 99.6%

• GPQA Diamond#37 of 144
  90.5%vendorReflection AI: Introducing Beam ↗

  Leader: GPT-6 Astra at 96.1%

• Humanity’s Last Exam#48 of 153
  36.2%vendorno toolsReflection AI: Introducing Beam ↗

  Leader: Claude Sonnet 5.5 at 64.5%

• CritPt#32 of 144
  16.3%vendorCritPt-AAReflection AI: Introducing Beam ↗

  Leader: Muse Wonder Full at 33%

Vision and long context

• AA Long Context Reasoning#45 of 144
  79.3%vendorvendor runReflection AI: Introducing Beam ↗

  Leader: Kimi K3 at 88.7%

Other published results