---
title: "LLM Context Window Sizes — Community Reference Table"
url: https://stacklist.com/card/6cfe3a46-9f6f-4bc9-969c-af14027e341d
source_url: "https://github.com/taylorwilsdon/llm-context-limits"
stack: https://stacklist.com/c/technology/stack/e16dcdcb-06b9-481c-aef9-40b6d73a5c8e
summary: "LLM Context Window Sizes Reference is a comprehensive guide documenting max context window lengths, input/output token limits, and feature compatibility for models from OpenAI, Anthropic, Qwen, Mistral, Deepseek, Llama, Phi, Gemini, and more. It covers API-driven model parameters including image/audio/video input support, tooling (MCP) support, and reasoning options, with practical tips for local deployment using Ollama and open-webui."
tags: "llm, context-window, token-limits, openai, api-reference, model-comparison, large-language-models"
key_entities: "OpenAI (organization), Anthropic (organization), Mistral (organization), Deepseek (organization), Google Gemini (organization), GPT-5 (technology), GPT-4.1 (technology), GPT-4o (technology), Ollama (technology), open-webui (technology), context window (concept), token limits (concept), KV cache (concept), flash attention (concept), MCP (technology), Qwen (technology), Llama (technology), Phi (technology)"
classification: "reference"
content_hash: "sha256:60938075e9f7152864cc3a378c6a110daad4bbd52e7438435de31d227d8d087a"
acp_version: "0.2"
token_counts_approximate: 2224
visibility: public
agent_accessible: true
status: "final"
---

# LLM Context Window Sizes — Community Reference Table

OpenAI, Anthropic, Qwen, Mistral, Deepseek, Llama, Phi, Gemini & More - API Max Context, Output Token Limits & Feature Compatibility The missing context limit & parameter support guide for local and hosted LLMs Since OpenAI won't just be cool and give us a max context and max output parameter in the OpenAI API-compatible models endpoint spec, I put together a quick reference for my own use that perhaps others can benefit from. This table represents the max current context window length, max input token, and max output token limits for OpenAI via API. This does not apply to ChatGPT through the UI. If anything looks wrong, please flag it or cut a PR to update, and I'll happily merge once confirmed accurate. Tip Are you using open-webui ? You can configure the max context window in a persistent manner under the Settings -> Models interface under advanced parameters. Warning Editor's note - if you don't utilize a k/v cache , setting the max context (even if you're not filling it up) will use up a ton of VRAM and potentially degrade performance. I strongly encourage running Ollama with Flash Attention enabled via OLLAMA_FLASH_ATTENTION=1 & set OLLAMA_KV_CACHE_TYPE=q8_0 (you can use a q4_0 quant but quality will degrade more) This table provides a quick reference to the key parameters of OpenAI's available API-driven models. These values apply to OpenAI's officially hosted API and may not match 3rd party providers. OpenAI API Model Reference Tooling references MCP support. Reasoning Option refers to the LLM API having an option to turn on thinking for the model. Core Generative Models Model Context Window (Tokens) Max Output Tokens Image Input Audio Input Video Input Tooling Support Reasoning Option Notes GPT-5.2 Pro 400,000 128,000 ✅ Yes ❌ No ❌ No ✅ Yes ✅ Yes Version of GPT-5.2 that produces smarter and more precise responses. GPT-5.2 400,000 128,000 ✅ Yes ❌ No ❌ No ✅ Yes ✅ Yes The best model for coding and agentic tasks across industries GPT-5.1 400,000 128,000 ✅ Yes ❌ No ❌ No ✅ Yes ✅ Yes The best model for coding and agentic tasks with configurable reasoning effort GPT-5.1 Codex Max 400,000 128,000 ✅ Yes ❌ No ❌ No ❌ No ✅ Yes Our most intelligent coding model optimized for long-horizon, agentic coding tasks. GPT-5.1 Codex 400,000 128,000 ✅ Yes ❌ No ❌ No ❌ No ✅ Yes A version of GPT-5.1 optimized for agentic coding in Codex. GPT-5 Codex 400,000 128,000 ✅ Yes ❌ No ❌ No ❌ No ✅ Yes A version of GPT-5 optimized for agentic coding in Codex GPT-5 400,000 128,000 ✅ Yes ❌ No ❌ No ✅ Yes ✅ Yes Previous intelligent reasoning model for coding and agentic tasks with configurable reasoning effort GPT-5 mini 400,000 128,000 ✅ Yes ❌ No ❌ No ✅ Yes ✅ Yes Faster, cost-efficient GPT-5 variant. GPT-5 nano 400,000 128,000 ✅ Yes ❌ No ❌ No ✅ Yes ✅ Yes Fastest, most cost-efficient GPT-5 variant. GPT-4.1 1,047,576 32,768 ✅ Yes ❌ No ❌ No ✅ Yes ❌ No Smartest non-reasoning model. GPT-4.1 mini 1,047,576 32,768 ✅ Yes ❌ No ❌ No ❌ No ❌ No Smaller, faster version of GPT-4.1. GPT-4.1 nano 1,047,576 32,768 ✅ Yes ❌ No ❌ No ❌ No ❌ No Fastest, most cost-efficient version of GPT-4.1. GPT-4o 128,000 16,384 ✅ Yes ❌ No ❌ No ❌ No ❌ No Omni-modal predecessor to GPT-5. GPT-4o mini 128,000 16,384 ✅ Yes ❌ No ❌ No ❌ No ❌ No Fast, affordable small model for focused tasks. GPT-4 8,192 8,192 ❌ No ❌ No ❌ No ❌ No ❌ No Older base GPT-4 (Text-only). GPT-4 Turbo 128,000 4,096 ✅ Yes ❌ No ❌ No ❌ No ❌ No An older high-intelligence GPT model, supports images. GPT-3.5 Turbo 16,385 4,096 ❌ No ❌ No ❌ No ❌ No ❌ No Legacy GPT model for chat and non-chat tasks. o1 200,000 100,000 ✅ Yes ❌ No ❌ No ❌ No ✅ Yes Previous full o-series reasoning model. o1 mini 128,000 65,536 ❌ No ❌ No ❌ No ❌ No ✅ Yes Small model alternative to o1 (Deprecated). o1 Preview 128,000 32,768 ❌ No ❌ No ❌ No ❌ No ✅ Yes Preview of our first o-series reasoning model (Deprecated). o1 pro 200,000 100,000 ✅ Yes ❌ No ❌ No ❌ No ✅ Yes Version of o1 with more compute. o3 200,000 100,000 ✅ Yes ❌ No ❌ No ❌ No ✅ Yes Reasoning model for complex tasks, succeeded by GPT-5. o3 mini 200,000 100,000 ❌ No ❌ No ❌ No ❌ No ✅ Yes A small model alternative to o3. o4 mini 200,000 100,000 ✅ Yes ❌ No ❌ No ❌ No ✅ Yes Fast, cost-efficient reasoning model, succeeded by GPT-5 mini. Specialized and Non-Generative Models Model Name Category Multimodal Capability Notes o3 deep research Deep Research Unknown Most powerful deep research model. GPT Audio Audio Processing Audio input/Text output For audio input/output via Chat Completions API. GPT-4o mini TTS Audio (TTS) Text input/Audio output Text-to-speech model. GPT-4o Transcribe Audio (STT) Audio input/Text output Speech-to-text model. DALL-E 3 Image Generation Text input/Image output Previous generation image generation model. GPT Image 1 Image Generation Text input/Image output State-of-the-art image generation model. GPT-OSS 120b Open-Weight Text Only Most powerful open-weight model, fits into an H100 GPU. GPT-OSS 20b Open-Weight Text Only Medium-sized open-weight model for low latency. Anthropic API Model Reference Model Context Window Max Output Tokens Supports Temperature? Supports Streaming? Vision Support? Claude 4.5 Opus 200k tokens 64k tokens ✅ Yes ✅ Yes ✅ Yes Claude 4.5 Sonnet 200k/1000k (extended) tokens 64k tokens ✅ Yes ✅ Yes ✅ Yes Claude 4.0 Sonnet 200k/1000k (extended) tokens 64k tokens ✅ Yes ✅ Yes ✅ Yes Claude 3.7 Sonnet 200k tokens 8k tokens (128k extended w/ output-128k-2025-02-19 header) ✅ Yes ✅ Yes ✅ Yes Claude 3.5 Sonnet 200k tokens 8k tokens ✅ Yes ✅ Yes ✅ Yes Claude 3.5 Haiku 200k tokens 8k tokens ✅ Yes ✅ Yes ❌ No Claude 3 Opus 200k tokens 4k tokens ✅ Yes ✅ Yes ✅ Yes Claude 3 Sonnet 200k tokens 4k tokens ✅ Yes ✅ Yes ✅ Yes Claude 3 Haiku 200k tokens 4k tokens ✅ Yes ✅ Yes ✅ Yes Training Data Cut-off: Claude 3.7 Sonnet: October 2024 Claude 3.5 Sonnet: April 2024 Claude 3.5 Haiku: July 2024 Claude 3 Opus: August 2023 Claude 3 Sonnet: August 2023 Claude 3 Haiku: August 2023 DeepSeek API Model Reference Through official DeepSeek API. Self-hosted supports 128k. Model Context Window Max CoT Tokens Max Output Tokens Supports Streaming? Vision Support? deepseek-chat (deepseek v3) 64k tokens - 8k tokens ✅ Yes ❌ No deepseek-reasoner (deepseek r1) 64k tokens 32K tokens 8k tokens ✅ Yes ❌ No Qwen API Model Reference Self-hosted maximums. Please note that you must configure your inference engine to these maximums, as the default (e.g., Ollama @ 2048 tokens) is generally much lower than the model maximum. Model Context Window Max Output Tokens Supports Streaming? Vision Support? qwen2.5-coder-32b 131,072 tokens 8k tokens ✅ Yes ❌ No qwen2.5-72b-instruct 131,072 tokens 8k tokens ✅ Yes ❌ No qwen2.5-3b 32k tokens (default, 128k possible) 8k tokens ✅ Yes ❌ No qwq 32k tokens 8k tokens ✅ Yes ❌ No Mistral API Model Reference Self-hosted maximums. Model Context Window Max Output Tokens Supports Streaming? Vision Support? Mistral-7B-Instruct-v0 32k tokens 4k tokens ✅ Yes ❌ No Mistral Medium 32k tokens 4k tokens ✅ Yes ❌ No Mistral Small 32k tokens 4k tokens ✅ Yes ❌ No Mistral Large 32k tokens 4k tokens ✅ Yes ❌ No Mistral Nemo 128k tokens 4k tokens ✅ Yes ❌ No Gemini API Model Reference Includes Gemini (hosted) and Gemma (self hosted). Model Context Window Max Output Tokens Supports Streaming? Vision Support? gemini-2.0-flash 1,048k tokens 8k tokens ✅ Yes ❌ No gemini-2.5-pro 1,048k tokens 64k tokens ✅ Yes ❌ No gemma-3 128k tokens Unclear tokens ✅ Yes ❌ No Other Model Reference Model Context Window Max Output Tokens Supports Streaming? Vision Support? Llama3.3:70b 131,072 tokens 2k tokens ✅ Yes ❌ No Phi4 16k tokens 16k tokens (*combined window - 16k total split between input & output) ✅ Yes ❌ No Phi4 16k tokens 16k tokens ✅ Yes ❌ No OpenAI API Model Endpoint Compatibility This table provides a reference for which models are compatible with various OpenAI API endpoints. Endpoint Compatible Models /v1/assistants All o-series, all GPT-4o (except chatgpt-4o-latest ), GPT-4o-mini, GPT-4, and GPT-3.5 Turbo models. The retrieval tool requires gpt-4-turbo-preview (and subsequent dated model releases) or gpt-3.5-turbo-1106 (and subsequent versions). /v1/audio/transcriptions whisper-1 /v1/audio/translations whisper-1 /v1/audio/speech tts-1 , tts-1-hd /v1/chat/completions All o-series, GPT-4o (except for Realtime preview), GPT-4o-mini, GPT-4, and GPT-3.5 Turbo models and their dated releases. chatgpt-4o-latest dynamic model. Fine-tuned versions of gpt-4o , gpt-4o-mini , gpt-4 , and gpt-3.5-turbo . /v1/completions (Legacy) gpt-3.5-turbo-instruct , babbage-002 , davinci-002 /v1/embeddings text-embedding-3-small , text-embedding-3-large , text-embedding-ada-002 /v1/fine_tuning/jobs gpt-4o , gpt-4o-mini , gpt-4 , gpt-3.5-turbo /v1/moderations text-moderation-stable , text-moderation-latest /v1/images/generations dall-e-2 , dall-e-3 /v1/realtime (beta) gpt-4o-realtime-preview , gpt-4o-realtime-preview-2024-10-01