How much does Beam cost?

Beam is published as open weights (Apache 2.0) — there is no per-token API price. You self-host the model and pay infrastructure cost (GPU rental and operating overhead). Effective cost on commodity 24-48GB GPUs typically lands at $0.10-$0.50 per million output tokens.

What is the context window of Beam?

Beam supports a context window of 1000K tokens (1,000,000 tokens) and can generate up to 33K output tokens per request.

What is Beam best for?

Beam is best for: Apache-2.0 large open model for self-hosted coding agents. It supports chat, function_calling, code, reasoning capabilities. It is open-source under the Apache 2.0 license.

Beam

Reflection AIopen-source NewOpen Source

Reflection AI's first model: a 501B-total / 23B-active sparse mixture-of-experts with 52 layers and a 1M-token effective context (256K longest used in reinforcement learning). Previewed 5 Oct 2026; weights released 16 Oct 2026 on Hugging Face under Apache 2.0 with BF16, FP8 and NVFP4 checkpoints — the permissive licence is its main selling point among large open models. Reflection's hosted API was in a free public beta with no published rate, so the row shows as self-host. Reflection reports (vendor-reported) Terminal-Bench 2.1 80.1, SWE-bench Verified 80.9 and AIME 2026 97.8; independent scoring is much weaker: Artificial Analysis Intelligence Index v5.0.1 gives it 18 (30 on the retired v4.3.2 scale), #19 of 26 open-weights models, though at only 53.7M output tokens, $0.28 per Index task and 287 tokens/s it is cheap and fast to run. Maximum output is not published; the figure shown is a placeholder.

Context Window

tokens

Max Output

33K

tokens

Input Price

Self-host

open weights

Output Price

Self-host

infra cost only

Speed

287

tokens/sec

Released

Oct 2026

2026-10-16

Blended Cost

—

self-host

Value Score

—

n/a self-host

Capabilities

ChatFunction CallingCode GenerationReasoning

Benchmarks

Quality Index

Price vs Quality

Context Window in Context

▸ Beam1M
GPT-4.11M
Gemini 2.5 Pro1M
Gemini 2.0 Flash1M
Llama 4 Maverick1M
GPT-5.51M
GPT-5.5 Pro1M
Claude Opus 4.71M

Bar length is logarithmic — each full step is 10× the tokens. Shown against the models with the nearest context windows, not the directory extremes.

Try in Playground Use via Swfte Connect Reflection AI Docs

Compare With

Open Source: Licensed under Apache 2.0

Non è ancora sul nostro gateway

Beam è a pesi aperti, quindi puoi eseguirlo tu stesso. Swfte Connect trasforma il tuo deployment in un endpoint gestito che parla la stessa API di ogni altro modello di questa directory.

La richiesta

curl https://api.swfte.com/agents/v2/gateway/chat/completions \
  -H "Authorization: Bearer $SWFTE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "reflection:beam-1.0",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Distribuisci questo modello con Swfte Connect

About Beam

Beam is a open-source AI model by Reflection AI, released on October 16, 2026. It supports a context window of 1000K tokens and can generate up to 33K output tokens.

Beam is published as open weights (Apache 2.0) for self-hosting. There is no per-token API price. Cost depends on your inference infrastructure: GPU rental, throughput per GPU, and operating overhead. For commodity 24-48GB GPUs the effective cost typically lands in the $0.10-$0.50 per million output tokens range, well below the cheapest hosted alternatives.

Beam is available as an open-source model under the Apache 2.0 license, meaning you can self-host it for predictable costs or use it through API providers like Swfte Connect.

Using Beam with Swfte

Access Beam through Swfte Connect, our unified LLM gateway. Connect gives you a single API for 50+ models, with automatic routing, cost optimization, and fallback handling. You can also try Beam in our AI Playground before integrating.