Reflection AI · Released Oct 5, 2026Beam 1.0 501B A23B
Sparse mixture-of-experts reasoning model with 23B active parameters for coding and agentic workflows.
- Capability
- —No ECI score yet
- Input
- $0.30per 1M tokens
- Output
- $1.00per 1M tokens
- Context
- 1.05M131K max output
What it costs
- Input
- $0.30per million tokens
- Output
- $1.00per million tokens
- Blended (3:1)
- $0.475cheaper than 62% of priced models
Typical monthly bills
| Monthly usage | Estimated cost |
|---|---|
| Side project2M input + 0.5M output tokens | $1.10 |
| Team assistant25M input + 5M output tokens | $12.50 |
| Production app250M input + 50M output tokens | $125.00 |
Median of 7 paid API providers on models.dev.
Independent benchmarks
Epoch AI has not published a Capabilities Index score for Beam 1.0 501B A23B yet.
The details
- Lab
- Reflection AI
- Released
- Oct 5, 2026
- Knowledge cutoff
- Jun 30, 2026
- Context window
- 1,048,576 tokens
- Max output
- 131,072 tokens
- Inputs
- Text
- Output
- Text
- Reasoning
- Always on
- Tool calling
- Yes
- Structured output
- No
- Weights
- Open weights · Apache-2.0
- Availability
- 12 API providerslisted on models.dev
What Reflection AI claims
Published by the lab at launch. Settings vary, so compare these only with care.
| Benchmark | Score | Setting | Source |
|---|---|---|---|
| DeepSWE v1.1 | 43.3 | BF16 | Source |
| SWE-Bench Pro v2 | 76.5 | BF16 | Source |
| SWE-Bench Pro v1 | 66.1 | BF16 | Source |
| Terminal-Bench v2.1 | 80.1 | BF16 | Source |
| SWE-Atlas Codebase QnA | 31.8 | BF16 | Source |
| SWE-Bench Multilingual | 78.3 | BF16 | Source |
| SWE-Bench Verified | 81.6 | BF16 | Source |
| AIME 2026 | 97.7 | BF16 | Source |
| Humanity's Last Exam | 36 | BF16; no tools | Source |
| Humanity's Last Exam | 51.1 | BF16; with tools | Source |
| SciCode | 49.2 | BF16 | Source |
| CritPt | 16.3 | BF16 | Source |
About Beam 1.0 501B A23B
How much does Beam 1.0 501B A23B cost?
Beam 1.0 501B A23B costs $0.30 input / $1.00 output per million tokens (median across 7 API providers). At a 3:1 input-to-output mix that is $0.475 per million tokens, cheaper than 62% of the 445 priced models we track.
What is the context window of Beam 1.0 501B A23B?
Beam 1.0 501B A23B accepts up to 1,048,576 tokens per request and can write up to 131,072 tokens in one response.
How good is Beam 1.0 501B A23B?
Epoch AI has not published a Capabilities Index score for Beam 1.0 501B A23B yet.
Is Beam 1.0 501B A23B open source?
Yes. Reflection AI publishes the weights under the Apache-2.0 licence, so it can be downloaded and self-hosted.
What inputs does Beam 1.0 501B A23B support?
Beam 1.0 501B A23B accepts text and replies in text. It is a reasoning model, supports tool calling.
When was Beam 1.0 501B A23B released?
Reflection AI released Beam 1.0 501B A23B on Oct 5, 2026. Its training data runs to Jun 30, 2026.
What are the best alternatives to Beam 1.0 501B A23B?
The closest current models from other labs on capability, price and release date are Ling 3.1 Flash (inclusionAI, $0.30 / $0.90), Perceptron Mk1.5 (Perceptron, $0.154 / $1.54), LongCat-2.5-Preview (Meituan (LongCat), $0.30 / $1.20) and MiMo-V2.6-Pro (Xiaomi, $0.435 / $0.87).