Context Engineering

Context window sizes by AI model

Context windows change often, and product limits can differ from API limits. This guide gives practical, source-backed defaults and shows how to think about the budget before you upload a large file.

By Riflet Updated 2026-09-08

Current practical defaults

Anthropic

Claude Fable 5.11,000,000 tokens: Claude Fable 5.1 supports a 1M-token context window and up to 128K output tokens. Anthropic's highest-capability model, built for demanding reasoning and long-horizon agentic work. List price: $10 per million input tokens and $50 per million output tokens.
Claude Opus 51,000,000 tokens: Claude Opus 5 supports a 1M-token context window and up to 128K output tokens via the standard Messages API. List price: $5 per million input tokens and $25 per million output tokens.
Claude Sonnet 51,000,000 tokens: Claude Sonnet 5 supports a 1M-token context window and up to 128K output tokens. List price: $2 per million input tokens and $10 per million output tokens.
Claude Haiku 4.5200,000 tokens: Claude Haiku 4.5 supports a 200K-token context window and up to 64K output tokens. List price: $1 per million input tokens and $5 per million output tokens.

OpenAI

GPT-6 Astra1,050,000 tokens: GPT-6 Astra supports a 1.05M-token context window, up to 922K input tokens and 128K output tokens. OpenAI's most capable model. Astra Pro is the same model served at a higher reasoning effort and shares these limits. List price: $10 per million input tokens and $50 per million output tokens.
GPT-5.6 Sol1,050,000 tokens: GPT-5.6 Sol supports a 1.05M-token context window and up to 128K output tokens. The highest-capability tier of the GPT-5.6 family. List price: $4 per million input tokens and $20 per million output tokens.
GPT-5.6 Terra1,050,000 tokens: GPT-5.6 Terra matches Sol's 1.05M-token context window and 128K output limit at a lower price. The balanced default. List price: $2 per million input tokens and $12 per million output tokens.
GPT-5.6 Luna1,050,000 tokens: GPT-5.6 Luna keeps the 1.05M-token context window and 128K output limit at the lowest cost. Good for high-volume context workflows. List price: $0.20 per million input tokens and $1.20 per million output tokens.

Google

Gemini 3.7 Flash1,048,576 tokens: Gemini 3.7 Flash supports a 1M-token context window and up to 64K output tokens. Google's current default model. List price: $0.75 per million input tokens and $3.75 per million output tokens.
Gemini 3.1 Pro1,048,576 tokens: Gemini 3.1 Pro supports a 1M-token context window and up to 64K output tokens. Still published as a preview model. List price: $2 per million input tokens and $12 per million output tokens.

Moonshot AI

Kimi K31,048,576 tokens: Kimi K3 supports a 1M-token context window. Output defaults to 128K tokens and can be raised to the full 1M. List price: $3 per million input tokens and $15 per million output tokens.

DeepSeek

DeepSeek V4 Pro1,000,000 tokens: DeepSeek V4 Pro supports a 1M-token context window and up to 384K output tokens. DeepSeek's flagship open-weight model. List price: $1.32 per million input tokens and $3.96 per million output tokens.
DeepSeek V4 Flash1,000,000 tokens: DeepSeek V4 Flash matches the Pro model's 1M-token context window and 384K output limit at lower cost. List price: $0.44 per million input tokens and $1.32 per million output tokens.

xAI

Grok 4.6500,000 tokens: Grok 4.6 documents a 500K-token context window. It is xAI's recommended general-purpose and coding model. List price: $2 per million input tokens and $6 per million output tokens.
Grok 4.31,000,000 tokens: Grok 4.3 documents a 1M-token context window, so it takes a larger bundle than the newer Grok 4.6. List price: $1.25 per million input tokens and $2.50 per million output tokens.

Alibaba

Qwen3.8-Max1,000,000 tokens: Qwen3.8-Max supports a 1M-token context window and up to 128K output tokens. Alibaba's flagship open-weight model. List price: $2 per million input tokens and $6 per million output tokens.

Mistral

Mistral Medium 3.5256,000 tokens: Mistral Medium 3.5 is Mistral's frontier-class model with a 256K-token context window. List price: $1.50 per million input tokens and $7.50 per million output tokens.
Mistral Large 3256,000 tokens: Mistral Large 3 is Mistral's open-weight general-purpose model with a 256K-token context window. List price: $0.50 per million input tokens and $1.50 per million output tokens.
Mistral Small 4256,000 tokens: Mistral Small 4 is a hybrid instruct, reasoning, and coding model with a 256K-token context window. Lower cost than Medium. List price: $0.15 per million input tokens and $0.60 per million output tokens.

Perplexity

Sonar128,000 tokens: Perplexity Sonar documents a 128K context length for web-grounded search queries. List price: $1 per million input tokens and $1 per million output tokens.
Sonar Pro200,000 tokens: Perplexity Sonar Pro documents a 200K context length and an 8K output limit for deeper research queries. List price: $3 per million input tokens and $15 per million output tokens.

Local

Local LLM32,768 tokens: Local-model limits depend on the model and runtime. The calculator default is editable.

Treat these as planning numbers, not a permanent spec. Vendors update models, plans, and product surfaces, so any serious workflow should keep the limits in one updateable place.

Input, output, and hidden overhead

A context window is shared by source material, instructions, conversation history, tool results, and the answer. If you fill the whole window with documents, the model has less room to respond.

Reasoning models may also reserve internal reasoning or output space. That is why a safe calculator should let you reserve answer room instead of assuming every token can be used for files.

How to budget a bundle

Start by estimating the source tokens. Reserve space for your instructions, follow-up question, and expected answer. Then trim the source bundle until the important material fits comfortably.

A smaller, curated bundle usually beats a maximum-size dump because the model can focus on the documents that actually support the task.

Why Riflet tracks this

Riflet shows token estimates while you select files, so token budgeting happens before export instead of after an AI rejects the upload or loses focus.

For teams, putting context limits in an updateable config also makes monthly model-limit updates simpler.

Build better AI context with Riflet

Collect the files, notes, websites, and docs that matter. Export one portable bundle for any AI.

Download Riflet free

Frequently asked questions

Is context window sizes by ai model only useful for developers?

No. The same context principles apply to research, writing, legal review, client work, product planning, and any workflow where the AI needs project-specific source material.

Does more context always produce better AI output?

No. Relevant, well-structured context usually beats a large dump of loosely related files. Large context windows make preparation more important, not less important.

Where does Riflet fit in this workflow?

Riflet is the assembly step. It helps gather source material, filter it, estimate the token budget, and export one portable file for the AI tool you choose.

Does Riflet replace RAG or AI project features?

Not always. RAG is useful for large retrieval systems. AI project features are useful inside one platform. Riflet is best when you want explicit, portable context you can inspect before using.