Skip to content

context-comparison.md

Latest commit

 

History

History
212 lines (162 loc) · 9.58 KB

context-comparison.md

File metadata and controls

212 lines (162 loc) · 9.58 KB

Context Window Comparison Report

Last Updated: 2026-02-06

Claude Code

Model Context Window Max Output Input $/M Output $/M Notes
Opus 4.6 200k (1M beta) 128k $5.00 $25.00 Latest model (Feb 5, 2026), adaptive thinking, first Opus with 1M context
Opus 4.5 200k 64k $5.00 $25.00 Premium model, complex reasoning, 67% lower cost than predecessor
Sonnet 4.5 200k (1M beta) 64k $3.00 $15.00 Recommended default, context awareness feature, best coding model
Haiku 4.5 200k 64k $1.00 $5.00 Fastest, context awareness feature, unbeatable for UI work

Legacy Models (Still Available):

  • Opus 4.5: 200k tokens, 64k max output
  • Opus 4.1: 200k tokens, 32k max output
  • Sonnet 4: 200k / 1M beta, 64k max output
  • Sonnet 3.7: 200k tokens, 64k / 128k beta max output
  • Opus 4: 200k tokens, 32k max output
  • Haiku 3: 200k tokens, 4k max output

Reference: Anthropic Models

Model Selection

  • During session: /model <alias>
  • At startup: claude --model <alias>
  • Environment: ANTHROPIC_MODEL=<alias>
  • Aliases: opus, sonnet, haiku, opusplan
  • Extended context: claude --betas context-1m-2025-08-07

Key Features

  • Opus 4.6 supports adaptive thinking with 4 effort levels (low, medium, high, max)
  • All 4.5 models support extended thinking (31,999 token budget by default)
  • 128k max output for Opus 4.6; 64k for other 4.5 models
  • Automatic context compaction at 75-92% utilization
  • Context awareness: Sonnet 4.5 and Haiku 4.5 track remaining context window throughout conversation
  • Context budget: 200K (standard), 500K (Enterprise), 1M (beta for tier 4 orgs)

Codex CLI

Model Context Window Max Output Input $/M Output $/M Notes
GPT-5.3-Codex 400k 128k TBD TBD Latest model (Feb 5, 2026), "Perfect Recall" attention, 25% faster
GPT-5.2-Codex 400k 128k $1.75 $14.00 Native compaction, reliable tool calling (Jan 14, 2026)
GPT-5.1-Codex-Max Multi-window (millions) N/A $1.25 $10.00 First model natively trained for compaction
GPT-5-Codex 200k N/A $1.25 $10.00 Standard model with 200K context window
codex-mini-latest ~200k N/A - - DEPRECATED - Removed Jan 16, 2026

Reference: OpenAI Models

Key Features

  • Context compaction is now a core feature (not experimental)
  • Automatically summarizes conversation history when approaching limits
  • Enables long-running coding tasks spanning millions of tokens
  • Model switching mid-session with /model command
  • ChatGPT account sign-in (no manual API key needed)
  • Agent Skills system production-ready

Cursor

Anthropic Models

Model Context Window Max Output Input $/M Output $/M Notes
Claude 4.6 Opus 200k (1M max) 64k $5.00 $25.00 Latest Opus model, deep reasoning
Claude 4.5 Sonnet 200k (1M max) 64k $3.00 $15.00 Premium, Agent/Thinking modes
Claude 4.5 Opus 200k 64k $5.00 $25.00 Deep reasoning
Claude 4.5 Haiku 200k 64k $1.00 $5.00 Fast and efficient
Claude 4 Sonnet 1M 1M 64k $3.00 $15.00 Max mode only, extended context
Claude 4 Sonnet 200k 64k $3.00 $15.00 Normal and Max modes
Claude 4 Opus 200k 64k $5.00 $25.00 Max mode only
Claude 3.7 Sonnet 200k 64k $3.00 $15.00 Normal and Max modes
Claude 3.5 Sonnet 200k 64k $3.00 $15.00 Fast and reliable
Claude 3.5 Haiku 60k 8k $0.80 $4.00 Speed-optimized

OpenAI Models

Model Context Window Max Output Input $/M Output $/M Notes
GPT-5.2 (all variants) 272k - $2.00 $8.00 Latest generation
GPT-5.2-Codex 272k - $1.75 $14.00 Code-optimized
GPT-5.1 Codex Max 200k (400k max) 128k $1.25 $10.00 Improved tool calls
GPT-5.1 High 200k (400k max) 128k $1.25 $10.00 Frontier coding model
GPT-5 Fast 272k - $1.25 $10.00 Speed-optimized
GPT-5 Mini 200k 128k $0.40 $1.60 Fast, lightweight option
GPT-5 Nano 200k 128k $0.15 $0.60 Ultra-fast for simple tasks
GPT 4.1 200k (1M max) 8k $2.00 $8.00 Normal and Max modes
GPT-4o 128k 16k $2.50 $10.00 All-purpose
GPT-4o mini 60k 8k $0.15 $0.60 Free (500/day)
o3 200k 8k $2.00 $8.00 Reasoning optimized
o4-mini 200k 8k $1.10 $4.40 Normal and Max modes
o3-mini 200k 8k $1.10 $4.40 Agent/Thinking modes
o1 200k 8k $15.00 $60.00 Thinking mode
o1 Mini 128k 8k $1.10 $4.40 Thinking mode

Google Models

Model Context Window Max Output Input $/M Output $/M Notes
Gemini 3 Pro Preview 200k (1M max) 8k $2.00 $12.00 Top performer on SWE-bench (76.2-78%)
Gemini 3 Pro Image Preview 200k (1M max) 8k $2.00 $12.00 Multimodal with image generation
Gemini 3 Flash 200k (1M max) 8k $0.50 $3.00 Fast inference
Gemini 2.5 Pro 200k (1M max) 8k $1.25 $10.00 Strong for design/debugging
Gemini 2.5 Flash 200k (1M max) 8k $0.30 $2.50 Fast inference, extended context

xAI Models

Model Context Window Max Output Input $/M Output $/M Notes
Grok 4 200k (2M max) - $3.00 $15.00 Real-time web search integration
Grok Code 256k - $2.00 $10.00 Code-optimized
Grok Code Fast 200k - $0.60 $4.00 Optimized for speed

DeepSeek Models

Model Context Window Max Output Input $/M Output $/M Notes
DeepSeek V3 64k-128k - $0.27 $1.10 Native support, hosted on US servers
DeepSeek V3.1 128k - $0.27 $1.10 Upgraded context from V3
DeepSeek R1 200k - $0.55 $2.19 Native support, free within Pro plan

Cursor Native Models

Model Context Window Max Output Input $/M Output $/M Notes
Composer 1 ~200k - N/A N/A 4x faster than comparable models, <30s response

Note: Cursor uses subscription pricing ($20/mo Pro, $40/mo Business). API prices shown above are the underlying provider rates, not what Cursor users pay directly.

Reference: Cursor Models

Context Mode Comparison

Mode Context Range Cost Speed
Normal 60k - 400k Standard Fast
Max 200k - 2M Premium Slower

Key Features

  • Dynamic context discovery reduces token usage by 46.9% for MCP tool calls
  • Automatic summarization when context window fills
  • Chat history preserved as files for context recovery
  • Efficient updates via Merkle tree-based change detection

Gemini CLI

Model Context Window Max Output Input $/M Output $/M Status
Gemini 3 Pro Preview 1M 65k $2.00 $12.00 Current flagship
Gemini 3 Flash Preview 1M 65k $0.50 $3.00 Balanced speed/intelligence - Launched Feb 4, 2026
Gemini 3 Pro Image Preview 65k 32k $2.00 $12.00 Image generation
Gemini 2.5 Pro 1M 65k $1.25 $10.00 Stable
Gemini 2.5 Flash 1M 65k $0.30 $2.50 Stable
Gemini 2.5 Flash-Lite 1M 65k $0.10 $0.40 Stable
Gemini 2.5 Flash Image 65k 32k $0.30 $2.50 Stable
Gemini 2.5 Flash Preview 1M 65k $0.30 $2.50 Experimental features (Preview)
Gemini 2.5 Flash TTS 8k 16k $0.30 $2.50 Text-to-speech (Preview)
Gemini 2.5 Flash Live 131k 8k $0.30 $2.50 Preview
Gemini 2.0 Flash 1M 8k $0.10 $0.40 Deprecated - March 31, 2026
Gemini 2.0 Flash-Lite 1M 8k $0.08 $0.30 Deprecated - March 31, 2026

Reference: Gemini API

Deprecation Timeline

Model Shutdown Date
Gemini 2.0 Flash March 31, 2026
Gemini 2.0 Flash-Lite March 31, 2026
Gemini 2.5 Pro June 17, 2026
Gemini 2.5 Flash June 17, 2026
Gemini 2.5 Flash-Lite July 22, 2026

Retired Models

Model Context Window Retirement Date
Gemini 1.5 Pro 2M Sept 24, 2025
Gemini 1.5 Flash 1M Sept 24, 2025

Note: The 2M token context window is no longer available. Maximum context is now 1M tokens.

Research Sources

Claude Code

Codex CLI

Cursor

Gemini CLI
⚡