whatmodel.sh / Beam 1.0 501B A23B
Beam 1.0 501B A23B
↗ share
+ contribute
Beam 1.0 501B A23B is an open-weights language model from Reflection, released October 2026. Its published size is 501.3B parameters (23.4B active per token) and a 1M-token context window. Licence: apache-2.0.
Catalogue identity notes
Final 16 October released checkpoint; BF16 lab results, Arena settings retained per result. Earlier 5 October prerelease scores excluded.
Model information sources
Score within the LMArena (text) comparison set only.
$ how it compares
The LLMs ranked either side of Beam 1.0 501B A23B, out of 882 with a score.
| # | model | score | where |
|---|---|---|---|
| 81 |  Grok 4.20 (0309 reasoning) | 71±1 | api $1.56/M |
| 82 |  GPT-5.4 mini | 70±1 | api $1.69/M |
| 83 | Beam 1.0 501B A23B | 70±1 | api price unknown |
| 84 |  Ling 3.0 Flash (124B-A5B) | 70±2 | api $0.090/M +local |
| 85 | GPT-5.1 | 70±1 | api $3.44/M |
$ what it's good at
agent tasks▁▁▃▆█▅▅▅▅▄▂▁ #54 of 310 * · 45% of its score · 10 tests
Beats Gemini 3.1 Pro (Preview), Muse Spark 1.1, Muse Spark 1.2 here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena Agent (October 2026 aggregate IPS) | -0.0435 IPS ✎ | ▁▁▁▃▃▃█▁▁▁▁▁ beats 13/52 | 45% |
| SWE-bench Verified lab | 81.6% ✎ | ▁▁▁▁▁▁▂▅█▆▁▁ beats 128/134 | — |
| SWE-bench Multilingual lab | 78.3% ✎ | ▁··▁▃▁▄▇█▂▂▁ beats 49/62 | — |
| Terminal-Bench 2.1 lab | 80.1% ✎ | ▁·▁·▁▁▄▃▅█▅█ beats 41/61 | — |
| Humanity's Last Exam with tools lab | 51.1% ✎ | ▂▁▁▁▄▂▃▅█▇▂▅ beats 32/56 | — |
| MCP-Atlas (public set) lab | 80.8% ✎ | ▂·▁▂▁·▁▃█▆▆▃ beats 37/38 | — |
| DeepSWE v1.1 (resolved) lab | 43.3% ✎ | ▂····▃▁▁▆███ beats 3/31 | — |
| τ³-bench Banking lab | 36.1% ✎ | ▂▂·▅█▂▃▃▃▆▂▅ beats 23/24 | — |
2 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| SWE-bench Pro lab | 66.1% ✎ | ▄▂▂▂▄▂▆█·▂·▃ beats 16/16 | — |
| BrowseComp with context management lab | 76.8% ✎ | ▄·▄▂·█▂·▆▄·▂ beats 6/15 | — |
coding▁▁▁▃▆▇█▇▅▄▃▁ #74 of 466 * · 35% of its score · 7 tests
Beats DeepSeek V4 Flash Vision (Exp), GPT-5.2, GPT-5 here, though they rank higher overall. Trails MiMo-V2.5 (310B-A15B), MiMo-V2.5-Pro (1T-A42B), GLM-5 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena WebDev (Code Arena) | 1436.9 ✎ | ▁▁▃▃▆█▄▇▅▅▁▁ beats 58/126 | 35% |
| SWE-bench Verified lab | 81.6% ✎ | ▁▁▁▁▁▁▂▅█▆▁▁ beats 128/134 | — |
| SWE-bench Multilingual lab | 78.3% ✎ | ▁··▁▃▁▄▇█▂▂▁ beats 49/62 | — |
| LiveCodeBench (version unspecified) lab | 88.3% ✎ | ▁···▁▁▁█▇▃▄▃ beats 38/38 | — |
| SciCode (level unspecified) lab | 49.2% ✎ | ▁····▃▄▄▅█▇▃ beats 36/37 | — |
| DeepSWE v1.1 (resolved) lab | 43.3% ✎ | ▂····▃▁▁▆███ beats 3/31 | — |
| SWE-bench Pro lab | 66.1% ✎ | ▄▂▂▂▄▂▆█·▂·▃ beats 16/16 | — |
chat & writing▁▁▁▃▆▄▆█▆▆▇▃ #73 of 398 * · 19% of its score · 1 test
Beats LongCat 2.0 (1.6T-A48B), GPT-5, Mistral Large 4 (public preview) here, though they rank higher overall. Trails Gemma 4 26B-A4B, Nemotron Cascade 2 30B-A3B, Gemma 4 31B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena (text) | 1435.9 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 294/382 | 19% |
mathnot ranked on this task yet · 1 test
web searchnot ranked on this task yet · 2 tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Humanity's Last Exam with tools lab | 51.1% ✎ | ▂▁▁▁▄▂▃▅█▇▂▅ beats 32/56 | — |
| BrowseComp with context management lab | 76.8% ✎ | ▄·▄▂·█▂·▆▄·▂ beats 6/15 | — |
other tests5 not in a task · 0.0% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GPQA Diamond lab | 90.6% ✎ | ▁▁▁▁▂▄▅▅█▇█▄ beats 276/315 | — |
| Humanity's Last Exam lab | 36.0% ✎ | ▁▁▁▄▅▆▆██▆▁▃ beats 86/108 | — |
| IFBench (aggregation unspecified) lab | 79.6% ✎ | ▃▇▂▄▅▅▂▄▄▆█▇ beats 76/83 | — |
| LongBench v2 lab | 67.5% ✎ | ▂·▁·▂▂▁▃▄▁█▄ beats 28/29 | — |
| AutomationBench (Public) lab | 39.3% ✎ | █·····▄····▄ beats 3/3 | — |
$ where to run it
Through an API
api · price unknown · 1M context · developers.reflection.ai ↗
Official weights and documentation: huggingface.co/reflection ↗
$ how the score was worked out · LMArena leaderboard dataset 99%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 LMArena leaderboard dataset | 3 | 99% |
| prior · pull toward the middle | 1% | |
| weighted average | = 70.3 |
not counted (17)
| Reflection Beam 1.0 model card | DeepSWE v1.1 (resolved) [correct] | 43.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | SWE-bench Pro [correct] | 66.1% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | Terminal-Bench 2.1 [correct] | 80.1% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | SWE-bench Multilingual [correct] | 78.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | SWE-bench Verified [correct] | 81.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | AIME 2026 [correct] | 97.7% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | Humanity's Last Exam [correct] | 36.0% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | Humanity's Last Exam with tools [correct] | 51.1% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | SciCode (level unspecified) [correct] | 49.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | GPQA Diamond [correct] | 90.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | LiveCodeBench (version unspecified) [correct] | 88.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | AutomationBench (Public) [correct] | 39.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | MCP-Atlas (public set) [correct] | 80.8% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | τ³-bench Banking [correct] | 36.1% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | BrowseComp with context management [correct] | 76.8% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | LongBench v2 [correct] | 67.5% | fewer than three models have this test yet, or this source ran it for this model alone |
| Reflection Beam 1.0 model card | IFBench (aggregation unspecified) [correct] | 79.6% | fewer than three models have this test yet, or this source ran it for this model alone |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json