AITOPAI热报
ModelsSelected

Reflection AI's Beam 1.0 Scores 23.5% on ARC-AGI-2 Verified

官方账号ARC Prize@arcprize
Post
View the post
Why we picked it

Cost per task is the notable figure: high reasoning trades just 3.5 points on ARC-AGI-2 for 43% savings, which matters for teams budgeting reasoning workloads. Researchers comparing frontier abstract-reasoning scores against models like GPT-5 or Gemini can pull the verified ARC Prize entry directly. The ARC-AGI-1 versus ARC-AGI-2 gap also offers a data point on how quickly reasoning models are closing the abstraction gap.

Beam 1.0 from Reflection AI posted 23.5% on ARC-AGI-2 (Verified) at US$0.61 per task, and 74.0% on ARC-AGI-1 at US$0.31 per task. Its best results came at the xhigh reasoning setting. The high setting scored 20.0% on ARC-AGI-2, close behind at roughly 43% lower cost. The numbers were published on 2 Jul 2026 by ARC Prize.

Original · ARC Prize

Beam 1.0 from @reflection_ai on ARC-AGI (Verified):

- ARC-AGI-2: 23.5%, $0.61/task - ARC-AGI-1: 74.0%, $0.31/task

Beam’s best scores came at xhigh reasoning, but high was close on ARC-AGI-2: 20.0% versus 23.5%, at about 43% lower cost. https://t.co/Et3fnkAPaw

#Reflection AI#Beam 1.0#ARC-AGI#reasoning models#benchmarks

Keep reading

Recent stories with the same tag or section
PapersUS vs ChinaSelected16:33
DeepSeek V4.1 RL environment counts still hard to compare

RL environment counting has become a reporting gap: different labs may describe the same training run with wildly different sandbox numbers, making cross-lab comparisons unreliable. Practitioners tracking frontier RL scaling should read both the DeepSeek paper and the Beam blogpost before citing either figure. The expectation that V4.1F-1017 outperforms Beam sets up a concrete head-to-head comparison to watch.

#DeepSeek#Reflection AI#RL environments#reasoning models
Teortaxes@teortaxesTexSource
IndustryMulti-sourceSelected13:28
NVIDIA AI apologises to Reflection AI over mistaken account tag

The mix-up hints that another account or company with the Reflection AI name is getting attention, which can matter for teams tracking the brand. Readers monitoring NVIDIA's partner communications should check the tagged handles before citing tweets as announcements. No regional angle exists; the tweet itself carried no substantive product news.

#NVIDIA#Reflection AI#Hugging Face
Event
官方账号NVIDIA AI@NVIDIAAI11 sourcesSource
ModelsUS vs ChinaSelected03:05
Reflection AI's Beam 1.0 Enters Agent Arena Top 15 Labs

A US lab landing a top open-weight spot in agentic benchmarks breaks the pattern of Chinese labs dominating the open WebDev leaderboard, where 24 of the top 25 models are Chinese. Teams choosing an Apache 2.0 model for agent or coding work now have a credible US alternative to Kimi K2.5 Thinking and GLM-5. The wide error bars mean its true standing relative to Qwen3.8 is less settled than the rank suggests.

#Beam 1.0#Reflection AI#open-source models#agents
lmarena.ai@lmarena_aiSource
ModelsBuildersSelected22:24
Reflection AI Releases Beam, First Open-Weight Model

Apache 2.0 weights with a 1M context window make Beam one of the few permissively licensed models built explicitly for agentic coding workloads. Developers can run it locally or via Ollama Cloud with zero data retention, and it plugs directly into Claude Code, Codex and OpenCode through Ollama's launcher commands. Teams evaluating open alternatives to closed coding models now have a new option to benchmark against their own agent pipelines.

#Reflection AI#Beam#Ollama#open-source models
ollama@ollamaSource
ModelsBuildersSelected22:21
Reflection AI releases Beam open weights under Apache 2.0

An Apache 2.0 release with NVFP4 quantised weights and a companion training harness is rare, since most open-weight drops stop at the checkpoint. Teams running inference on NVIDIA hardware can take the DFlash2 drafter and FP8/NVFP4 variants directly into production stacks without negotiating licences. Engineers evaluating open models against Llama or Qwen releases should benchmark Beam on their own workloads.

#Reflection AI#Beam#open-source models#Hugging Face
官方账号Hugging Face@huggingfaceSource
ModelsUS vs ChinaSelected22:06
Reflection AI's Beam-1.0-501B-A23B Ranks #12 in Agent Arena

24 of the top 25 open models on Code Arena: WebDev come from Chinese labs, and Beam 1.0 is the lone US exception — a notable shift in the open-weights competitive landscape. Builders evaluating Apache 2.0 agentic models can compare Beam directly against Qwen3.8, Kimi K2.5 Thinking and GLM-5 once the full weights drop this month. Its 23B active parameter design targets reasoning efficiency for agent workloads.

#Reflection AI#Beam 1.0#open-source models#agents
lmarena.ai@lmarena_aiSource