Beam 1.0 is Reflection's open-weight AI model, released 16/10/2026. Reflection AI's launch post presents Beam as a text-only sparse mixture-of-experts model with 501B total and 23B active parameters, 23.8T pretraining tokens and an Apache 2.0 weight release promised for later in October 2026. The post says it matches GLM-5.2 on advanced reasoning with 3 to 4 times less inference compute (FLOPs approximated as 2 x active parameters x mean generated tokens). Tables reflect an update dated 27 October 2026 and the page gives no evaluation settings, so rows carry no setting. Entries the page marks NR are omitted from the comparison columns.
Visit Beam 1.0AI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
How does Beam 1.0 (Beam 501B) score on independent benchmarks?
Beam 1.0 (Beam 501B) ranks #27 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 17.7, as measured on 22 November 2026.
Where does Beam 1.0 (Beam 501B) rank against other models?
How good is Beam 1.0 (Beam 501B) at agentic coding?
How does Beam 1.0 (Beam 501B) do on individual tests?
| Evaluation (run by Artificial Analysis) | Beam 1.0 (Beam 501B) | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 3% | 67.2% |
| FrontierCode 2.0 | 5.5% | 48.8% |
| Humanity's Last Exam | 34.7% | 74.3% |
| GDPval-AA (Elo) | 1177 | 1866 |
How has Beam 1.0 (Beam 501B)'s score changed over time?
We have tracked Beam 1.0 (Beam 501B) on Intelligence Index v5.0.1 since 22/11/2026. A point is added each day the score, coding index or price changes, so this chart fills in over time.
What does the Beam 1.0 (Beam 501B) system card report?
Reflection AI's launch post presents Beam as a text-only sparse mixture-of-experts model with 501B total and 23B active parameters, 23.8T pretraining tokens and an Apache 2.0 weight release promised for later in October 2026. The post says it matches GLM-5.2 on advanced reasoning with 3 to 4 times less inference compute (FLOPs approximated as 2 x active parameters x mean generated tokens). Tables reflect an update dated 27 October 2026 and the page gives no evaluation settings, so rows carry no setting. Entries the page marks NR are omitted from the comparison columns.
These are Reflection AI's own published results (self-reported, with the vendor's harness and settings), so they are not directly comparable with other labs' numbers. Source: Introducing Beam (tables updated 27 October 2026) (05/10/2026). Last checked 22/11/2026.
Coding
| Benchmark | Beam 1.0 (Beam 501B) | GLM 5.2 | GLM 5.3 | Kimi K3 |
|---|---|---|---|---|
| DeepSWE v1.1 | 44.4% | 44% | 61% | 68% |
| SWE Bench Pro v2-Hard | 77.2% | n/a | 84.3% | 88.2% |
| SWE Bench Pro v1 | 65.5% | 62.1% | n/a | n/a |
| Terminal Bench v2.1 | 80.1% | 81% | 88.2% | 88.3% |
| SWE Atlas Codebase QnA | 34.6% | n/a | 61% | 68% |
| SWEBench Multilingual | 78% | n/a | n/a | n/a |
| SWEBench Verified | 80.9% | n/a | n/a | n/a |
| SciCode | 49.7% | n/a | 59% | 58.7% |
Reasoning & maths
| Benchmark | Beam 1.0 (Beam 501B) | GLM 5.2 | GLM 5.3 | Kimi K3 |
|---|---|---|---|---|
| AIME 2026 | 97.8% | 99.2% | n/a | n/a |
| CritPt AA | 16.3% | 20.9% | 19.1% | 23.4% |
Knowledge & science
| Benchmark | Beam 1.0 (Beam 501B) | GLM 5.2 | GLM 5.3 | Kimi K3 |
|---|---|---|---|---|
| HLE (no tools) | 36.2% | 40.5% | 42.3% | 46.9% |
| HLE (with tools) | 51.2% | 54.7% | 62.5% | n/a |
| GPQA Diamond | 90.5% | 91.2% | 91.7% | 93.5% |
| AA Omniscience (index), public split onlyReflection's own results on the public split | 13 points | n/a | 22.6 | n/a |
Agents & tool use
| Benchmark | Beam 1.0 (Beam 501B) | GLM 5.2 | GLM 5.3 | Kimi K3 |
|---|---|---|---|---|
| AutomationBench public | 37% | 26.2% | 48.2% | 46.7% |
| MCP Atlas | 78.7% | 77.8% | 84.2% | 82.3% |
| tau3 banking | 38% | 37.1% | n/a | 37.1% |
| BrowseComp with context management | 77.4% | n/a | n/a | 91.2% |
| DeepSearchQA with context management | 80.1% | n/a | n/a | 95% |
Long context
| Benchmark | Beam 1.0 (Beam 501B) | GLM 5.2 | GLM 5.3 | Kimi K3 |
|---|---|---|---|---|
| AA-LCR | 79.3% | 78.3% | 79.7% | 88.7% |
| LongBench v2 | 66.9% | 64% | n/a | n/a |
Other
| Benchmark | Beam 1.0 (Beam 501B) | GLM 5.2 | GLM 5.3 | Kimi K3 |
|---|---|---|---|---|
| IFBench | 79.7% | 73.3% | n/a | n/a |
What did safety testing find for Beam 1.0 (Beam 501B)?
Safety level: Not yet published; the post says Beam is undergoing final red-teaming and evaluations
- Alignment uses deliberative alignment, with safety policy incorporated into reasoning and an adversarially built dataset, plus safety RL across single-turn, multi-turn, jailbreak and agentic scenarios.
- A second model was trained with separate SFT and RL pipelines, and two teachers were merged using multi-teacher on-policy distillation.
- Safety evaluation results are to appear in the technical report, and internal safety evaluations are to be open-sourced; no safety scores are on the page.
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page. Full analysis: our Beam 1.0 (Beam 501B) deep dive.
Reflection Model Timeline
RL rollouts to 256K tokens; mid-training extends effective context to 1M tokens context
Specifications
Related Tools
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.
GPT-6.1 Sol
GPT-6.1 Sol, launched at OpenAI DevDay on 29 September 2026, is OpenAI's mid-tier workhorse model. OpenAI says it delivers nearly the same intelligence as GPT-6 Astra at one-fifth of Astra's token price: $2 input / $10 output per million tokens with cached input cut to $0.10. It scores 75.2% on DeepSWE v1.1 (high effort) and 71.4% on OSWorld 2.0, has a 1.05M-token context window and 128K max output, and is available in the API plus ChatGPT Work and Codex.
