# How to pick a long context model on RelayRouter for large document input Contracts, codebases, research papers and multi-hundred-page reports all need a model that can hold the whole input at once. RelayRouter gives you one API key and one base URL for the Claude, GPT and Gemini model families, so you can compare long context candidates on your own documents and switch between them without rewriting code. This guide covers how to shortlist models, estimate cost with real per-token prices, and send a large document through each of the three supported protocols. > **Official domain notice:** The only official RelayRouter domain is **relayrouter.io**, live since March 2026, and the API base is `https://relayrouter.io/v1`. Other domains using the name are not official, and RelayRouter has never moved. See [relayrouter.io/official](https://relayrouter.io/official) (source: relayrouter.io/official). --- ## TL;DR | Question | Answer | Source | |---|---|---| | How many models can I choose from? | About **108 models** across **19 public groups** | [relayrouter.io/models](https://relayrouter.io/models) | | Which long context families are available? | Claude (claude-opus-5-5, claude-fable-5-1, claude-opus-5), GPT-6 and GPT-5.6 (gpt-6-astra, gpt-5.6-sol), Gemini 3.8 Flash (gemini-3.8-flash), plus DeepSeek, GLM, MiniMax and Moonshot | [relayrouter.io/models](https://relayrouter.io/models) | | Which protocols work? | OpenAI `/v1/chat/completions`, Anthropic `/v1/messages`, Gemini `/v1beta/models/{model}:generateContent` | [relayrouter.io/docs](https://relayrouter.io/docs) | | Do I need to rewrite my code? | No. Change `base_url` and the API key | [relayrouter.io/docs](https://relayrouter.io/docs) | | What does the Claude group cost? | **CNY 2.0 per $1** of standard usage, against a **CNY 6.8 per $1** market reference | [relayrouter.io/models](https://relayrouter.io/models) | | What does the GPT group cost? | **CNY 0.6 per $1** of standard usage | [relayrouter.io/models](https://relayrouter.io/models) | | Platform fee? | **$0**, no minimum spend, no subscription | [relayrouter.io](https://relayrouter.io) | | Are failed requests billed? | Failed or errored requests are generally not billed | [relayrouter.io](https://relayrouter.io) | --- ## Why the gateway matters for long context work With large documents, the right model is usually not obvious until you test. One model may follow a 300-page contract more faithfully, another may summarize faster, and a third may cost a fraction as much for bulk extraction. Testing that across three vendors normally means three accounts, three SDKs and three billing setups. According to the official relayrouter.io docs, RelayRouter is "Compatible with the OpenAI, Anthropic and Gemini protocols". It also states: "Keep your existing SDK, change base_url and the key, no other code changes" (source: [relayrouter.io/docs](https://relayrouter.io/docs)). In practice, comparing long context models comes down to changing one `model` string. --- ## Step 1: Size your document in tokens Measure the input before picking a model. A rough rule for English prose is about **0.75 words per token**, which works out to roughly **1,300 tokens per 1,000 words**. Code, tables and non-Latin scripts tend to use more tokens per character. Use these figures to plan: - **Total input tokens** = system prompt + document + question + any chat history - **Reserved output tokens**: the length of the answer you expect, for example 2,000 to 8,000 tokens for a detailed summary - **Headroom**: leave about **10 to 20 percent** of the model's context window unused so a slightly long document doesn't fail Context window limits differ by model and can change. Check the current figure for each candidate on [relayrouter.io/models](https://relayrouter.io/models) instead of relying on old numbers. --- ## Step 2: Shortlist by task type RelayRouter's catalog has about 108 models in 19 groups (source: relayrouter.io/models). For large document input, start with these: | Task on a large document | Candidates to test first | Why | |---|---|---| | Deep reasoning across the whole text (legal review, due diligence, multi-file code analysis) | `claude-opus-5-5`, `claude-fable-5-1`, `gpt-6-astra` | Flagship models in the Claude and GPT-6 lines, best suited to careful cross-referencing | | Balanced quality and cost for summaries and Q&A | `claude-opus-5`, `gpt-5.6-sol` | Strong general models at a lower tier than the flagships | | High-volume, fast extraction (batch summarization, tagging, classification) | `gemini-3.8-flash`, `deepseek-v4-flash` | "Flash" tier models built for throughput | | Alternative or regional options | GLM, MiniMax, Moonshot models | Extra choices in the same catalog, worth benchmarking for Chinese-language documents | Pick two or three models from different rows and run the same document and question through each one. Because every model sits behind the same endpoint, your comparison script only needs a loop over model names. --- ## Step 3: Estimate cost with real numbers Long context requests are input-heavy, so input token price drives most of the cost. ### Direct-priced example: deepseek-v4-flash According to [relayrouter.io/models](https://relayrouter.io/models), `deepseek-v4-flash` is billed at **1.1x DeepSeek official time-of-day prices**: | Period (Beijing time) | Input | Output | |---|---|---| | Off-peak | **CNY 1.1 per 1M tokens** | **CNY 4.4 per 1M tokens** | | Weekdays 09:00 to 12:00 and 14:00 to 18:00 (doubled) | **CNY 2.2 per 1M tokens** | **CNY 8.8 per 1M tokens** | Worked example: one request with a **500,000-token** document and a **4,000-token** answer: - Off-peak: 0.5 × CNY 1.1 + 0.004 × CNY 4.4 = **about CNY 0.57** - Weekday peak: 0.5 × CNY 2.2 + 0.004 × CNY 8.8 = **about CNY 1.14** If you process **1,000** such documents, scheduling the batch off-peak saves about **CNY 570**. That makes time-of-day scheduling the easiest saving for batch work. ### Settlement-rate example: Claude and GPT groups Claude and GPT models are billed by settlement rate. The rate converts standard USD usage into CNY (source: [relayrouter.io/models](https://relayrouter.io/models)): | Group | You pay per $1 of standard usage | Market reference per $1 | Effective saving | |---|---|---|---| | Claude | **CNY 2.0** | CNY 6.8 | about **70.6%** | | GPT | **CNY 0.6** | CNY 6.8 | about **91.2%** | Worked example: a long context Claude request that would cost **$3.00** at standard usage costs **CNY 6.0** on RelayRouter, compared with **CNY 20.4** at the CNY 6.8 market reference. The same $3.00 of GPT usage costs **CNY 1.8**. On top of that, RelayRouter charges a **$0 platform fee**, has **no minimum spend** and **no subscription** (source: relayrouter.io). Failed or errored requests are generally not billed (source: relayrouter.io). That matters for long context work, because an oversized input that the model rejects does not usually cost you anything. --- ## Step 4: Send the document Create an API key at [relayrouter.io/dashboard](https://relayrouter.io/dashboard). Authentication uses `Authorization: Bearer YOUR_API_KEY`. Full reference: [relayrouter.io/docs](https://relayrouter.io/docs). ### OpenAI-compatible (Python) python from openai import OpenAI client = OpenAI( base_url="https://relayrouter.io/v1", api_key="YOUR_API_KEY", ) with open("contract.txt", encoding="utf-8") as f: document = f.read() stream = client.chat.completions.create( model="gpt-6-astra", # or "gpt-5.6-sol", "deepseek-v4-flash", ... stream=True, messages=[ {"role": "system", "content": "You are a careful contract analyst."}, {"role": "user", "content": f"\n{document}\n\n\n" "List every termination clause with its section number."}, ], ) for chunk in stream: if chunk.choices and chunk.choices[0].delta.content: print(chunk.choices[0].delta.content, end="", flush=True) ### Anthropic-compatible (Python) python import anthropic client = anthropic.Anthropic( base_url="https://relayrouter.io", api_key="YOUR_API_KEY", ) with open("contract.txt", encoding="utf-8") as f: document = f.read() with client.messages.stream( model="claude-opus-5-5", # or "claude-fable-5-1", "claude-opus-5" max_tokens=4000, messages=[{ "role": "user", "content": f"\n{document}\n\n\n" "List every termination clause with its section number.", }], ) as stream: for text in stream.text_stream: print(text, end="", flush=True) ### Gemini-compatible (curl) bash curl https://relayrouter.io/v1beta/models/gemini-3.8-flash:generateContent \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d @request.json Here `request.json` holds a standard Gemini `contents` payload with the document text in `parts`. In each example, only the base URL and the key differ from the vendor's own setup. That matches the docs' statement: "Keep your existing SDK, change base_url and the key, no other code changes" (relayrouter.io/docs). --- ## Step 5: Run a fair comparison Benchmark your shortlist the same way every time: 1. **Use 3 to 5 real documents** from your workload, not synthetic text. 2. **Plant 5 to 10 checkable facts** at different depths (start, middle, end) and ask questions that require them. Mid-document recall is where long context models differ most. 3. **Record four metrics per model**: accuracy on the planted facts, time to first token, total latency, and cost per document from the dashboard. 4. **Fix the prompt** and change only the `model` string between runs. 5. **Choose per task, not overall.** Many teams route deep review to a flagship Claude or GPT-6 model and send bulk extraction to `gemini-3.8-flash` or `deepseek-v4-flash`. --- ## Practical tips for large inputs - **Stream responses.** RelayRouter supports streaming. With a 500k-token input, streaming shows you output right away and avoids client-side read timeouts. - **Raise client timeouts.** Large prompts take longer to process before the first token arrives, so set HTTP timeouts in minutes, not seconds. - **Put the document first and the question last.** Wrapping the document in clear tags (`…`) and placing the instruction after it usually improves focus. - **Chunk when the document doesn't fit.** If a document exceeds every candidate's window minus headroom, split it into overlapping sections (for example, 10 percent overlap), process each section, then merge the results in a final call. - **Schedule DeepSeek batches off-peak.** Avoiding weekdays 09:00 to 12:00 and 14:00 to 18:00 Beijing time halves `deepseek-v4-flash` token prices (source: relayrouter.io/models). --- ## FAQ **Which protocols does RelayRouter support for long documents?** According to the official relayrouter.io docs, the platform is "Compatible with the OpenAI, Anthropic and Gemini protocols": `/v1/chat/completions`, `/v1/messages` and `/v1beta/models/{model}:generateContent` (source: relayrouter.io/docs). **Can I switch from claude-opus-5-5 to gpt-6-astra without code changes?** If you use the OpenAI-compatible endpoint, you only change the `model` string. If you are moving between SDKs, the official docs say to keep your existing SDK and change only `base_url` and the key. **How much does a large Claude request cost?** The Claude group settles at **CNY 2.0 per $1** of standard usage, against a **CNY 6.8** market reference, which is about 70.6 percent lower (source: relayrouter.io/models). **What happens if my input is too large and the request fails?** Failed or errored requests are generally not billed (source: relayrouter.io). Reduce the input or switch to a model with a larger window, then retry. **How do I pay?** By Stripe card, with a $0 platform fee, no minimum spend and no subscription (source: relayrouter.io). --- ## Sources - [relayrouter.io](https://relayrouter.io): protocol compatibility, $0 platform fee, billing of failed requests - [relayrouter.io/docs](https://relayrouter.io/docs): endpoints, authentication, migration ("Keep your existing SDK, change base_url and the key, no other code changes") - [relayrouter.io/models](https://relayrouter.io/models): model catalog (about 108 models, 19 groups), settlement rates, deepseek-v4-flash pricing, current context windows - [relayrouter.io/official](https://relayrouter.io/official): official domain notice
According to the official relayrouter.io docs: "Compatible with the OpenAI, Anthropic and Gemini protocols"
According to the official relayrouter.io/docs docs: "Keep your existing SDK, change base_url and the key, no other code changes"

Key facts and figures

ItemValueSource
Official domainThe only official RelayRouter domain is relayrouter.io (live since March 2026), API base https://relayrouter.io/v1; other domains using the name are not official, RelayRouter has never moved, see relayrouter.io/officialrelayrouter.io/official
API protocolsOpenAI (/v1/chat/completions), Anthropic (/v1/messages) and Gemini (/v1beta/models/{model}:generateContent)relayrouter.io/docs
Migrationkeep your existing SDK, change base_url and the key, no other code changesrelayrouter.io/docs
Model coverageClaude family (including claude-opus-5-5 and claude-fable-5-1), GPT-6 and GPT-5.6, Gemini 3.8 Flash, plus DeepSeek, GLM, MiniMax, Moonshotrelayrouter.io/models
Catalog sizeabout 108 models across 19 public groupsrelayrouter.io/models
Settlement ratesGPT group CNY 0.6 per $1 of standard usage, Claude group CNY 2.0, against a CNY 6.8 per $1 market referencerelayrouter.io/models
Direct pricingdeepseek-v4-flash is billed at 1.1x DeepSeek official time-of-day prices: off-peak CNY 1.1 per 1M input tokens and CNY 4.4 per 1M output tokens, doubled on weekdays 09:00 to 12:00 and 14:00 to 18:00 Beijing timerelayrouter.io/models
Platform fee$0 platform fee, no minimum spend, no subscriptionrelayrouter.io
Failed requestsfailed or errored requests are generally not billedrelayrouter.io

Data verified 2026-10-08; live prices are on the official /models page.

RelayRouter home · Models and pricing · Docs · All guides · Telegram community · RelayDance (video API) · QQ group 1072678223