benchrPricing options
| Option | Cost |
|---|---|
| Self-hosted (Meta weights) | Infrastructure only |
| Together.ai input | ~$0.11/1M |
| Together.ai output | ~$0.34/1M |
| Context window | 10,000,000 tokens |
What fits inside a 10M context window
Ten million tokens is approximately 7,500 pages of text. To put this in engineering terms: a 1M-line codebase with documentation may fit, as may a decade of customer-support transcripts or a large book collection. For comparison, Gemini 3.1 Pro lists a 1M-token maximum, so Scout's stated maximum is 10× larger. Actual usable capacity depends on provider limits, tokenization, and deployment memory.
Practically, a software company could target a single context containing an 800K-line production codebase plus tests and documentation, subject to tokenization and the serving stack's maximum. Compared with a 1M-token ceiling, the stated 10M maximum reduces the need for chunking on some very large workloads.
Scout vs Maverick: the context-quality tradeoff
Llama 4 Maverick scores higher on the cited reasoning and SWE-bench figures, while Scout emphasizes context length and throughput. If a task fits within 1M tokens, compare Maverick's task quality with Scout's lower quoted API rate. For requests above 1M, Scout is one long-context candidate, but verify competing offerings and the exact maximum exposed by each provider.
Hardware for self-hosting
Scout has 109B total parameters. Weight storage alone is roughly 218GB at BF16 and 54.5GB at ideal 4-bit packing, before quantization metadata, runtime buffers, and KV cache. A suitable Int4 checkpoint may fit its weights on an 80GB accelerator, but that does not mean the advertised 10M-token context fits there: context memory, concurrency, cache precision, and the serving implementation are separate constraints. Benchmark the exact checkpoint and target length; many deployments expose a smaller context cap.
Cost scenarios
At 5M input + 1M output per month (long-document processing): approximately $0.55 + $0.34 = ~$0.89/month via Together.ai. Gemini 3.1 Pro at the same volume, assuming each request stays at or below its 200K base-price threshold, is $10 + $12 = $22/month — about 25× higher. Price alone does not establish task quality. If an individual request exceeds Gemini 3.1 Pro's 1M maximum, compare Scout with other long-context offerings and verify the supported limit before committing.
Use-case fit
Best for: Full-repository code analysis; multi-year document archive processing; long-running agent sessions that must maintain complete context without summarization; workflows that require a verified provider limit above 1M tokens; cost-sensitive long-context retrieval where chunking introduces coherence loss.
Skip if: Your tasks fit in 1M context and reasoning quality is more important than context ceiling — Llama 4 Maverick or Gemini 3.1 Pro are stronger reasoning models. Also skip for multimodal (image) tasks — Scout is text-only.
Decision checklist
Identify your actual p95 context length requirement. If it's under 500K tokens, compare Llama 4 Maverick's 1M context and benchmark profile with Scout's price and throughput. Consider Scout when tasks regularly exceed 1M tokens and its task-specific quality is acceptable.
Verify provider support before building around the full 10M window. Self-hosting does not guarantee that ceiling either: calculate weight memory, KV cache, runtime overhead, and concurrency for the exact serving stack, then reproduce the target length under load.