Skip to content

context-comparison.md

Latest commit

 

History

History
271 lines (211 loc) · 14.1 KB

context-comparison.md

File metadata and controls

271 lines (211 loc) · 14.1 KB

Context Window Comparison Report

Last Updated: 2026-05-02

Claude Code

Model Context Window Max Output Input $/M Output $/M Notes
Opus 4.7 1M (GA) 128k $5.00 $25.00 Latest flagship (Apr 16, 2026), 1M GA, adaptive thinking, new tokenizer (~555k words/1M)
Opus 4.6 1M (GA) 128k $5.00 $25.00 Released Feb 5, 2026; 1M GA since Mar 13, 2026; adaptive thinking
Sonnet 4.6 1M (GA) 64k $3.00 $15.00 1M GA since Mar 13, 2026; default for Pro/Enterprise/API plans
Sonnet 4.5 200k 64k $3.00 $15.00 Legacy (active); no 1M context
Haiku 4.5 200k 64k $1.00 $5.00 Fastest; no 1M context support; no adaptive thinking
Opus 4.5 200k 64k $5.00 $25.00 Legacy (active)
Opus 4.1 200k 32k $5.00 $25.00 Legacy (active)

Reference: Anthropic Models

Model Selection

  • During session: /model <alias>
  • At startup: claude --model <alias>
  • Environment: ANTHROPIC_MODEL=<alias>
  • Aliases: opus / best (→ Opus 4.7), sonnet (→ Sonnet 4.6), haiku (→ Haiku 4.5), opusplan
  • Explicit 1M variants: opus[1m], sonnet[1m]
  • Bedrock/Vertex/Foundry: opus resolves to Opus 4.6, sonnet resolves to Sonnet 4.5
  • Disable 1M: CLAUDE_CODE_DISABLE_1M_CONTEXT=1

Key Features

  • 1M context is GA (no longer beta) for Opus 4.7, Opus 4.6, and Sonnet 4.6 as of March 13, 2026
  • 300k output tokens (Beta) on Message Batches API for Opus 4.7 / Opus 4.6 / Sonnet 4.6 via output-300k-2026-03-24 header
  • Adaptive thinking: Opus 4.7, Opus 4.6, Sonnet 4.6 — dynamically allocates reasoning per step
  • Effort levels: Opus 4.7 supports low, medium, high, xhigh, max; Opus 4.6 / Sonnet 4.6 support low, medium, high, max
  • opusplan uses standard 200k context during plan phase (no automatic 1M upgrade)

Plan-Level 1M Context Availability

Plan Opus 1M Sonnet 1M
Max, Team, Enterprise Included with subscription Requires extra usage
Pro Requires extra usage Requires extra usage
API / pay-as-you-go Full access Full access

Deprecations and Retirements

Model Status Deprecated Retirement Replacement
claude-opus-4-20250514 Deprecated Apr 14, 2026 Jun 15, 2026 claude-opus-4-7
claude-sonnet-4-20250514 Deprecated Apr 14, 2026 Jun 15, 2026 claude-sonnet-4-6
claude-3-haiku-20240307 Retired Feb 19, 2026 Apr 20, 2026 claude-haiku-4-5
claude-3-5-haiku-20241022 Retired Dec 19, 2025 Feb 19, 2026 claude-haiku-4-5
claude-3-7-sonnet-20250219 Retired Oct 28, 2025 Feb 19, 2026 claude-sonnet-4-6
claude-3-5-sonnet-* Retired Aug 13, 2025 Oct 28, 2025 claude-sonnet-4-6
claude-3-opus-20240229 Retired Jun 30, 2025 Jan 5, 2026 claude-opus-4-7

Note: The entire Claude 3.x generation (Opus 3, Sonnet 3.5, Haiku 3, Haiku 3.5) is fully retired as of 2026.

Codex CLI

Model Context Window Max Output Input $/M Output $/M Notes
gpt-5.5 1.05M (API) / ~400k (Codex) 128k TBD TBD Latest frontier (Apr 23, 2026); Codex caps at 400k; known issue #19319 may reduce effective ctx to ~258k
gpt-5.4 1.05M 128k $2.50 (doubles >272k) TBD Flagship; experimental 1M in Codex via model_context_window config
gpt-5.4-mini 400k 128k TBD TBD Default mini model
gpt-5.3-codex 400k 128k TBD TBD Default coding model (Feb 2026)
gpt-5.3-codex-spark 128k N/A N/A N/A Research preview; ChatGPT Pro only; >1,000 tokens/sec
gpt-5.2 400k 128k $2.00 $8.00 Previous frontier; still supported
gpt-5-codex 400k 128k $1.25 $10.00 DEPRECATED — retires Jul 23, 2026
gpt-5.1-codex 400k 128k $1.25 $10.00 DEPRECATED — retires Jul 23, 2026
gpt-5.2-codex 400k 128k $1.75 $14.00 DEPRECATED — retires Jul 23, 2026
codex-mini-latest 200k — — — RETIRED Feb 12, 2026
o4-mini 200k — $1.10 $4.40 DEPRECATED — shutdown Oct 23, 2026
o3-mini 200k — — — RETIRED Feb 23, 2026
o3 200k — — — DEPRECATED

Reference: Codex Models

Key Features

  • Auto-compaction triggers at 95% of model_context_window by default; in CLI v0.100.0+ user-defined limit is silently clamped to 90%
  • Server-side compaction via OpenAI Responses API; emits compaction output item in stream
  • After compaction, Codex re-reads up to 5 recently edited files (50k token budget, 5k per file)
  • Configurable via model_auto_compact_token_limit in config.toml
  • ChatGPT account sign-in (no manual API key needed)
  • Standalone /responses/compact endpoint available for stateless compaction

Configuration Example

model = "gpt-5.5"
model_context_window = 400000
model_auto_compact_token_limit = 350000
tool_output_token_limit = 50000

Deprecation Timeline

Model Shutdown Date Replacement
codex-mini-latest Feb 12, 2026 (retired) gpt-5.3-codex
o3-mini Feb 23, 2026 (retired) o3 (also deprecated)
gpt-5-codex Jul 23, 2026 gpt-5.3-codex
gpt-5.1-codex Jul 23, 2026 gpt-5.3-codex
gpt-5.2-codex Jul 23, 2026 gpt-5.3-codex
o3-deep-research Jul 23, 2026 gpt-5.4-pro
o4-mini Oct 23, 2026 gpt-5-mini

Cursor

Anthropic Models

Model Context Window Max Output Input $/M Output $/M Notes
Claude 4.7 Opus 200k (1M Max) 128k $5.00 $25.00 Current flagship Anthropic
Claude 4.6 Opus 200k (1M Max) 128k $5.00 $25.00 Current; deep reasoning
Claude 4.6 Sonnet 200k (1M Max) 64k $3.00 $15.00 Current; 1M Max Mode
Claude 4.5 Opus 200k (200k Max) 64k $5.00 $25.00 Legacy; Max Mode capped at 200k
Claude 4.5 Sonnet 200k (1M Max) 64k $3.00 $15.00 Legacy (active)
Claude 4.5 Haiku 200k (200k Max) 64k $1.00 $5.00 Hidden by default
Claude 4 Sonnet 1M 200k (1M Max) 64k $3.00 $15.00 Hidden; 2x cost over 200k tokens
Claude 4 Sonnet 200k (200k Max) 64k $3.00 $15.00 Hidden by default

OpenAI Models

Model Context Window Max Output Input $/M Output $/M Notes
GPT-5.5 200k (1M Max) 128k TBD TBD Newest OpenAI in Cursor
GPT-5.4 200k (1M Max) 128k $2.50 TBD Briefly defaulted to Max Mode for legacy billing
GPT-5.4 Mini 200k (1M Max) 128k TBD TBD Mini variant
GPT-5.4 Nano 200k (1M Max) 128k TBD TBD Smallest variant
GPT-5.3 Codex 272k (272k Max) 128k TBD TBD Coding-optimized
GPT-5.2 272k (272k Max) 128k $2.00 $8.00 Active
GPT-5.2 Codex 272k (272k Max) 128k $1.75 $14.00 Active
GPT-5.1 Codex 272k (272k Max) 128k $1.25 $10.00 Active
GPT-5 272k (272k Max) 128k $1.25 $10.00 Active
GPT-5 Mini 200k (200k Max) 128k $0.40 $1.60 Active

Google Models

Model Context Window Max Output Input $/M Output $/M Notes
Gemini 3.1 Pro 200k (1M Max) 64k $2.00 $12.00 Current Google leader
Gemini 3 Pro 200k (1M Max) 64k $2.00 $12.00 Current
Gemini 3 Pro Image Preview 200k (1M Max) 32k $2.00 $12.00 Multimodal preview
Gemini 3 Flash 200k (1M Max) 64k $0.50 $3.00 Fast inference
Gemini 2.5 Flash 200k (1M Max) 65k $0.30 $2.50 Legacy (active)

xAI Models

Model Context Window Max Output Input $/M Output $/M Notes
Grok 4.20 200k (2M Max) — $3.00 $15.00 Primary xAI model; replaces Grok Code
Grok Code Fast 1 256k — $0.60 $4.00 Active; may be hidden by default
Grok Code (original) 256k — $2.00 $10.00 Removed from default list; re-add via custom model ID

Cursor Native Models (Composer)

Model Context Window Max Output Input $/M Output $/M Notes
Composer 2 200k — $0.50 (Std) / $1.50 (Fast) — Default in Auto mode (Mar 19, 2026); built on Kimi K2.5 base
Composer 1.5 200k — N/A N/A Hidden by default; adds Thinking support
Composer 1 200k — N/A N/A First-gen; hidden by default

Other Providers

Model Context Window Max Output Input $/M Output $/M Notes
Kimi K2.5 (Moonshot) 200k (1M Max) — TBD TBD Active
DeepSeek V4 Pro 1M (1M Max) — TBD TBD Custom model / API key required
DeepSeek V4 Flash 1M (1M Max) — TBD TBD Custom model / API key required

Note: Cursor uses subscription pricing (Pro / Business). Composer models have unlimited access on paid plans in Auto mode. Third-party models draw from the monthly credit pool. API prices shown are underlying provider rates, not what Cursor users pay directly.

Reference: Cursor Models

Normal vs Max Mode (Major Behavioral Change — Early 2026)

  • 200k is now the default context for all models — previously Claude Sonnet required Max Mode to reach 200k.
  • Max Mode is reserved exclusively for 1M+ context access (e.g., Gemini 3 Pro at 1M, Grok 4.20 at 2M).
  • Max Mode uses token-based pricing (consumes credits faster than request-based).
  • Tab Autocomplete (Cursor-small) uses a separate ~8k context model, not configurable.
  • Confirmed by Lee Robinson (Cursor): "Max mode is only needed for 1M context windows."

Key Features

  • Dynamic context discovery reduces token usage for MCP tool calls
  • Automatic summarization when context window fills
  • Cursor 3.2 (Apr 24, 2026) is current — multi-tasking agents, workspaces, background agent tasks

Gemini CLI

Model Context Window Max Output Input $/M Output $/M Status
Gemini 3.1 Pro Preview 1M 64k $2.00 $12.00 Current flagship; supersedes gemini-3-pro-preview
Gemini 3 Flash Preview 1M 64k $0.50 $3.00 Available in Gemini CLI since Dec 2025
Gemini 3.1 Flash-Lite Preview 1M 64k TBD TBD New (Mar 3, 2026); cost-efficient tier
Gemini 3.1 Flash Image Preview 128k 32k TBD TBD Image generation preview
Gemini 3 Pro Image Preview 65k 32k $2.00 $12.00 Multimodal with image generation
Gemini 2.5 Pro 1M 65k $1.25 $10.00 Stable / Active — guaranteed not before Oct 16, 2026
Gemini 2.5 Flash 1M 65k $0.30 $2.50 Stable / Active — guaranteed not before Oct 16, 2026
Gemini 2.5 Flash-Lite 1M 65k $0.10 $0.40 Stable / Active
Gemini 2.0 Flash 1M 8k $0.10 $0.40 Deprecated — EOL Jun 1, 2026
Gemini 2.0 Flash-Lite 1M 8k $0.08 $0.30 Deprecated — EOL Jun 1, 2026
Gemini 3 Pro Preview (original) 1M 64k — — DISCONTINUED Mar 26, 2026 — redirects to 3.1

Reference: Gemini API

Gemini CLI Notes

  • Gemini CLI v0.40.1 (Apr 30, 2026) is current stable
  • Intelligent auto-routing: simple queries → Flash, complex tasks → Pro
  • Manual override: /model command or -m flag (e.g., gemini -m gemini-2.5-flash)
  • Gemma 4 experimental local-inference support added in v0.41.0-preview builds

Deprecation Timeline

Model Shutdown Date
gemini-3-pro-preview (original) March 26, 2026 (discontinued)
gemini-2.5-flash-lite-preview-09-2025 March 31, 2026 (discontinued)
gemini-robotics-er-1.5-preview April 30, 2026 (shut down)
Gemini 2.0 Flash June 1, 2026 (EOL)
Gemini 2.0 Flash-Lite June 1, 2026 (EOL)
Gemini 2.5 Pro Not before October 16, 2026
Gemini 2.5 Flash Not before October 16, 2026

Retired Models

Model Context Window Retirement Date
text-embedding-004 — January 14, 2026
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 in any active Gemini model. Maximum context is now 1M tokens for all current-gen Gemini Pro and Flash models.

Research Sources

Claude Code

Codex CLI

Cursor

Gemini CLI