What should I know about Beam?

Reflection AI announced Beam on October 5, 2026: a text-only sparse mixture-of-experts model with 501B total parameters and 23B active per token. Early-access beta for selected users through a waitlist; the documented API model ID is Beam-501B-A23B. Reasoning is mandatory, with medium as the default effort. Reflection developer documentation specifies a 262,144-token combined input/output context window and 131,072-token maximum output. Limits may change during beta. Weights, technical report, model card, and developer artifacts are promised later in October under Apache 2.0; weights are not yet released. The current weights flag reflects present availability. The launch benchmark table is provider-reported. Configuration, split, and source-policy compatibility must be verified before using these scores in AI IQ; no launch-table scores are imported. No current standard hosted token pricing, paired task costs/token usage, or raw speed measurements are published in the checked official sources.

What is the IQ of Beam?

Beam does not yet have an estimated IQ. AI IQ publishes a composite score once a model has source-backed benchmark results in at least two of its seven capability dimensions.

No — Beam is a proprietary model. Its weights are not publicly available, so it is used through hosted APIs and products.

When was Beam released?

Beam was publicly released on October 5, 2026.

How large is the context window of Beam?

Beam supports a context window of 262,144 tokens — the amount of text it can consider in a single request.

Reflection

Beam

Public AI IQ model profile with source-backed benchmark results and derived ranking context.

IQ

n/a

EQ

n/a

Effective Cost

n/a

IQ Rank

n/a

Release and Capabilities

Public release	2026-10-05
Modalities	text
Context window	262,144 tokens

Restricted early-access beta for selected users; waitlist at platform.reflection.ai. API model ID: Beam-501B-A23B.

Publisher-reported capabilities

Reflection AI announced Beam on October 5, 2026, for coding, reasoning, and agentic workloads.
Sparse mixture-of-experts architecture with 501B total parameters and 23B active per token.
Text-only model with configurable reasoning effort; the beta API documents a 256K context window.

Limitations and availability caveats

Access is restricted and beta behavior and limits may change.
Weights and an Apache 2.0 release, technical report, and model card are planned for later in October; they are not currently released.
Launch benchmark scores are self-reported and are not used as verified AI IQ measurements.
Pricing, paired task costs/token usage, and raw inference speed remain unverified.

Official release and model references

Measured Coverage and Estimates

Beam has 0 of 39 tracked IQ benchmark inputs across 0 of 7 dimensions. Raw benchmark tables contain source-backed results only. Derived IQ combines all six dimensions using the shared scoring methodology, including estimates where direct coverage is incomplete.

Benchmark versions, evaluation harnesses, and coverage affect comparisons; a missing result is not a score of zero. See the scoring methodology for estimation and uncertainty limits.

IQ Dimensions

Dimension	IQ	Coverage
Abstract Reasoning	99	0/5
Mathematical Reasoning	n/a	0/5
Academic Reasoning	100	0/6
Programmatic Reasoning	n/a	0/6
Computer Use	n/a	0/5
Reliability	n/a	0/6
Knowledge Work	n/a	0/6

Benchmark Results

Benchmark	Score	Dimension

Compare

Frequently Asked Questions

What should I know about Beam?: Reflection AI announced Beam on October 5, 2026: a text-only sparse mixture-of-experts model with 501B total parameters and 23B active per token. Early-access beta for selected users through a waitlist; the documented API model ID is Beam-501B-A23B. Reasoning is mandatory, with medium as the default effort. Reflection developer documentation specifies a 262,144-token combined input/output context window and 131,072-token maximum output. Limits may change during beta. Weights, technical report, model card, and developer artifacts are promised later in October under Apache 2.0; weights are not yet released. The current weights flag reflects present availability. The launch benchmark table is provider-reported. Configuration, split, and source-policy compatibility must be verified before using these scores in AI IQ; no launch-table scores are imported. No current standard hosted token pricing, paired task costs/token usage, or raw speed measurements are published in the checked official sources.
What is the IQ of Beam?: Beam does not yet have an estimated IQ. AI IQ publishes a composite score once a model has source-backed benchmark results in at least two of its seven capability dimensions.
Is Beam open source?: No — Beam is a proprietary model. Its weights are not publicly available, so it is used through hosted APIs and products.
When was Beam released?: Beam was publicly released on October 5, 2026.
How large is the context window of Beam?: Beam supports a context window of 262,144 tokens — the amount of text it can consider in a single request.