A comparison of GPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro using vendor-published benchmarks, context limits and pricing. Plan and API prices are dated September 2026.
ChatGPT, Claude, and Gemini are the three most prominent general-purpose AI assistants. This comparison covers the flagship models each vendor shipped in February and March 2026: OpenAI's GPT-5.4 (built-in computer use and a 1.05M-token API context window), Anthropic's Claude Opus 4.6 (80.8% on SWE-bench Verified), and Google's Gemini 3.1 Pro (94.3% on GPQA Diamond, per Google DeepMind's model card ↗).
Last updated: September 2026. Benchmark figures apply only to the model versions named. As of September 2026, newer models have shipped, including OpenAI's GPT-5.5 ↗ (April 2026) and Anthropic's Claude Opus 5.5 ↗ (September 22, 2026), while Gemini 3.1 Pro remains Google's current Pro model. Prices below are dated separately.
Claude Opus 4.6 posted 80.8% on SWE-bench Verified and ships with the Claude Code CLI. Gemini 3.1 Pro offers a 1M-token context window and scored 94.3% on GPQA Diamond. GPT-5.4 added built-in computer use for agentic workflows.
| Use Case | Strong option | Why |
|---|---|---|
| Coding & Development | Claude (Opus 4.6 + Claude Code) | 80.8% SWE-bench Verified, Claude Code CLI for terminal workflows |
| Research & Analysis | Gemini 3.1 Pro | 1M context, 94.3% GPQA Diamond, deep Google integration |
| Creative Writing | Claude Opus 4.6 | Long 128K output limit for long documents |
| Agentic Workflows | ChatGPT (GPT-5.4) | Built-in computer use, multi-step task automation |
| Best Value | Gemini | Free tier with Gemini Flash and some 3.1 Pro access, $19.99/mo AI Pro plan (as of September 2026) |
| Enterprise/Teams | ChatGPT | Mature ecosystem, Business and Enterprise plans, Codex for async dev work |
GPT-5.4 launched March 5, 2026 with built-in computer use and a 1.05M-token API context window. Claude Opus 4.6 launched February 5, 2026. Gemini 3.1 Pro was released in preview on February 19, 2026.
GPT-5.4 introduced built-in computer use capabilities. The model interprets screenshots, operates browsers, and issues keyboard commands to complete tasks across applications.
Key specifications, per OpenAI's API documentation ↗:
1.05M token context window (API)
128K max output tokens
Long-context pricing — prompts over 272K input tokens are billed at 2x input and 1.5x output
Computer use built-in — native screen interaction for agentic tasks
GPT-5.4 is available on paid ChatGPT plans, not the free tier. GPT-5.4 Pro is a separate, higher-cost variant.
Claude Opus 4.6 was released February 5, 2026. On Google DeepMind's published comparison table, it posted the highest SWE-bench Verified score of the models listed (80.8%).
Opus 4.6 specifications, per Anthropic's announcement ↗:
80.8% on SWE-bench Verified (as reported in Google DeepMind's Gemini 3.1 Pro comparison table)
200K context window (1M in beta on the Claude Developer Platform) with 128K max output tokens
Adaptive thinking — Claude decides how much extended thinking a task needs
Compaction — the API can summarize older context for longer-running tasks
Agent teams in Claude Code — multiple agents can work on a task together
Claude Sonnet 4.6 scored 79.6% on SWE-bench Verified in the same table and costs $3/$15 per million tokens versus Opus 4.6's $5/$25. Anthropic reports that users preferred Sonnet 4.6 over Opus 4.5 59% of the time ↗ in Claude Code.
Gemini 3.1 Pro was released in preview on February 19, 2026. Google DeepMind's model card reports:
94.3% on GPQA Diamond — the highest score in Google's comparison table
80.6% on SWE-bench Verified — 0.2 points behind Claude Opus 4.6
77.1% on ARC-AGI-2 — more than double Gemini 3 Pro's 31.1%
1M token context window with up to 64K output tokens
Multimodal input — text, images, audio, video, and PDFs in a single prompt
The 1M-token window lets users load long technical specifications or large codebases in one prompt without chunking.
Claude Opus 4.6 leads SWE-bench Verified by a narrow margin. Gemini 3.1 Pro posts the highest GPQA Diamond and ARC-AGI-2 scores in Google's table. GPT-5.4 has the largest API context window and built-in computer use.
| Feature | ChatGPT (GPT-5.4) | Claude (Opus 4.6) | Gemini (3.1 Pro) |
|---|---|---|---|
| Model | GPT-5.4 (Mar 2026) | Claude Opus 4.6 (Feb 2026) | Gemini 3.1 Pro (Feb 2026) |
| Context Window | 1.05M (API) | 200K (1M beta) | 1M |
| Max Output | 128K tokens | 128K tokens | 64K tokens |
| SWE-bench Verified | — | 80.8% | 80.6% |
| GPQA Diamond | — | 91.3% | 94.3% |
| ARC-AGI-2 | — | 68.8% | 77.1% |
| Image Generation | Built-in | No native generation | Nano Banana (built-in) |
| Computer Use | Built-in | Via API tool | Via separate Computer Use model |
| Code Execution | Yes (sandbox) | Yes | Yes (sandbox) |
| Web Search | Built-in | Built-in | Built-in (Google Search) |
| Voice Mode | Advanced Voice | Voice support | Gemini Live |
| Coding Agent | Codex | Claude Code CLI | Gemini CLI |
Benchmark cells marked "—" are not included in Google DeepMind's comparison table, which compared Gemini 3.1 Pro against Opus 4.6 and GPT-5.2 rather than GPT-5.4.
Claude Code runs in the terminal and IDEs and passed $2.5B in run-rate revenue by February 2026. Codex handles async task delegation in cloud sandboxes. Gemini CLI offers a free tier of 1,000 requests per day.
Anthropic reported in February 2026 that Claude Code's run-rate revenue had grown to over $2.5 billion ↗. The tool operates in terminals, reads project structures, writes code, runs tests, and handles git workflows without leaving the command line.
Claude Code capabilities:
Agentic search — scans the codebase for context without manual file selection
Subagents and agent teams — split exploration and implementation across parallel agents
MCP integration — connects to external tools such as Google Drive, Jira, and Slack through MCP servers
Full terminal access — runs builds, executes tests, manages git, handles CLI operations
VS Code and JetBrains extensions — same agent functionality inside IDEs
Opus 4.6 scored 80.8% on SWE-bench Verified, a benchmark of real GitHub issues. Sonnet 4.6 at 79.6% offers similar coding capability at a lower price.
Codex is built for task delegation. Users describe requirements and Codex works in cloud sandbox environments preloaded with their repositories.
Codex features:
Autonomous operation on multi-step tasks with progress updates
Cloud sandbox execution — isolated environments with test harnesses, linters, type checkers
Verifiable evidence — citations of terminal logs and test outputs for step tracing
Agent skills support — reusable instruction bundles for repeatable tasks
Parallel operation — worktrees let multiple agents work across a project
Interactive mode allows mid-task questions, approach discussions, and solution steering during code generation.
Some developers use Claude Code and Codex as complementary tools:
Claude Code generates — interactive implementation with local codebase context
Codex reviews — code review in a cloud sandbox with test execution
Claude Code iterates — fixes based on Codex feedback
A common split is Codex for async batch work and Claude Code for interactive development.
Gemini CLI runs on Gemini 3 models and, per its GitHub README ↗, offers a free tier of 60 requests per minute and 1,000 requests per day with a personal Google account (as of September 2026). It includes Google Search grounding, file operations, and shell commands.
Gemini CLI is a solid free option for individual developers and smaller tasks. Teams comparing it with Claude Code or Codex should trial all three on their own codebase.
In Google DeepMind's comparison table, Gemini 3.1 Pro leads GPQA Diamond (94.3%) and ARC-AGI-2 (77.1%), and Claude Opus 4.6 leads SWE-bench Verified (80.8%).
| Benchmark | GPT-5.2 | Claude Opus 4.6 | Gemini 3.1 Pro | What It Tests |
|---|---|---|---|---|
| GPQA Diamond | 92.4% | 91.3% | 94.3% | PhD-level science (physics, chemistry, biology) |
| ARC-AGI-2 | 52.9% | 68.8% | 77.1% | Abstract reasoning and pattern recognition |
Source: Google DeepMind's Gemini 3.1 Pro model card (February 2026). OpenAI's newer GPT-5.4 is not included in that table.
| Benchmark | GPT-5.2 | Claude Opus 4.6 | Gemini 3.1 Pro | What It Tests |
|---|---|---|---|---|
| SWE-bench Verified | 80.0% | 80.8% | 80.6% | Resolving real GitHub issues |
| SWE-bench Verified (Sonnet 4.6) | — | 79.6% | — | Mid-tier model comparison |
The top three scores sit within one percentage point of each other, so differences in tooling and workflow matter as much as the benchmark gap.
Human-preference rankings such as the LMArena leaderboard ↗ change frequently as new models are added. Check the live leaderboard for current positions rather than relying on a snapshot.
As of September 2026, all three offer a standard tier around $20/month. ChatGPT Pro costs $100 or $200/month, Claude Max starts at $100/month, and Google AI Ultra starts at $99.99/month. All three have free tiers.
| Plan | ChatGPT | Claude | Gemini |
|---|---|---|---|
| Free | Limited access to OpenAI's default model | Chat, web search, file creation | Gemini 3.6 Flash, varying access to 3.1 Pro |
| Standard | Plus: $20/mo | Pro: $20/mo ($17/mo billed annually) | AI Pro: $19.99/mo |
| Power User | Pro: $100 or $200/mo | Max: from $100/mo (5x or 20x Pro usage) | AI Ultra: $99.99 or $199.99/mo |
| What Standard Gets You | Thinking models, image generation, voice, higher limits | Claude Code, more usage, more models | Higher Gemini usage, Deep Research, video generation |
| Model | Input | Output | Notes |
|---|---|---|---|
| GPT-5.4 | $2.50 | $15.00 | 1.05M context, 2x input / 1.5x output above 272K |
| Claude Opus 4.6 | $5.00 | $25.00 | $10/$37.50 above 200K (1M beta) |
| Claude Sonnet 4.6 | $3.00 | $15.00 | Lower-cost Claude option for coding |
| Gemini 3.1 Pro | $2.00 | $12.00 | $4/$18 above 200K tokens |
| Tool | Access | Cost model |
|---|---|---|
| Claude Code | Claude Pro ($20/mo) or Max (from $100/mo) | Included in plan, or API usage |
| Codex | ChatGPT Plus ($20/mo) or Pro ($100–200/mo) | Included in plan, or API usage |
| Gemini CLI | Free (1,000 requests/day) | Free tier, or paid API key |
Gemini CLI's free tier covers many individual projects. Claude Pro at $20/mo includes Claude Code. ChatGPT Pro adds higher Codex limits for heavy users.
Choose ChatGPT for agentic automation and team environments. Choose Claude for daily coding and long-form writing. Choose Gemini for large-document processing and Google ecosystem integration.
Need agentic automation — GPT-5.4's built-in computer use operates browsers, apps, and multi-step workflows
Want async coding delegation — Codex handles fire-and-forget tasks in cloud sandboxes
Work in a team environment — ChatGPT's Business and Enterprise plans offer admin controls, SSO, and compliance features
Need image generation alongside conversation — built into ChatGPT without context switching
Value ecosystem breadth — custom GPTs and wide third-party integration support
Are a developer who codes daily — Claude Code CLI is built for terminal-based development
Need long-form writing — 128K output tokens support long documents in one pass
Work with complex, nuanced tasks — adaptive thinking scales reasoning to task difficulty
Want strong coding benchmark results — 80.8% on SWE-bench Verified
Care about AI safety policy — Anthropic publishes system cards and a constitution for its models
Need to process large documents — the 1M context window handles long codebases, videos, and PDFs without chunking
Are deep in the Google ecosystem — Gmail, Docs, Drive, and Workspace integration
Want a generous free tier — free users get Gemini Flash plus varying access to 3.1 Pro
Need strong reasoning scores — 94.3% GPQA Diamond and 77.1% ARC-AGI-2 are the highest in Google's comparison table
Work with multimodal input — text, images, audio, video, and PDFs in one prompt
A multi-tool configuration some professionals use:
Claude Code (Claude Pro, $20/mo) — primary coding tool for development work
ChatGPT Pro ($200/mo) — Codex for async task delegation and code review, plus agentic workflows
Google AI Pro ($19.99/mo) — long-document analysis, research with large context, Google Workspace integration
Total: about $240/month at September 2026 prices. Most individual users will not need all three.
Single choice for developers: Claude Pro at $20/mo, which bundles Claude Code with the chat app. For specialized voice and audio workflows, ElevenLabs covers that vertical.
"Best" depends on use case. GPT-5.4 brought built-in computer use and a 1.05M-token API context window. Claude Opus 4.6 led SWE-bench Verified at 80.8% in Google DeepMind's comparison table. Gemini 3.1 Pro led GPQA Diamond (94.3%). All three vendors have since released newer models, so check each vendor's current lineup.
On SWE-bench Verified, Claude Opus 4.6 scored 80.8% in Google DeepMind's comparison table, ahead of GPT-5.2 at 80.0%; OpenAI's newer GPT-5.4 was not in that table. Claude Code passed $2.5B in run-rate revenue by February 2026, per Anthropic, and offers terminal integration, subagents, and MCP support. Codex is built for async task delegation. Many developers use both: Claude Code for implementation, Codex for review.
Gemini 3.1 Pro leads specific benchmarks in Google's table: 94.3% GPQA Diamond (PhD-level science), 77.1% ARC-AGI-2 (abstract reasoning), and 80.6% SWE-bench Verified. It offers a 1M context window and a generous free tier. For research and document analysis, it is a strong choice.
As of September 2026, all three offer a standard tier at about $20/month: ChatGPT Plus ($20/mo), Claude Pro ($20/mo), Google AI Pro ($19.99/mo). Power-user tiers: ChatGPT Pro at $100 or $200/mo, Claude Max from $100/mo, and Google AI Ultra at $99.99 or $199.99/mo. ChatGPT also offers a lower-cost Go plan. For API usage, Gemini 3.1 Pro costs $2/$12 per MTok, GPT-5.4 costs $2.50/$15, and Claude Opus 4.6 costs $5/$25.
Yes. A common workflow uses Claude Code for interactive implementation with local codebase context and Codex for code review and async task delegation in cloud sandboxes with built-in test execution.
Among these three models, GPT-5.4 has the largest API context window at 1.05M tokens. Gemini 3.1 Pro supports 1M tokens. Claude Opus 4.6 supports 200K tokens as standard, with a 1M beta on the Claude Developer Platform.
The three-way AI race in early 2026 had no single "best" answer. Each platform carved out genuine strengths.
GPT-5.4 is a strong general-purpose agent. Built-in computer use, a broad ecosystem, and mature team plans make it a safe all-around choice, especially for teams and enterprises.
Claude Opus 4.6 led SWE-bench Verified in Google DeepMind's table, and Claude Code became one of the most widely adopted AI coding tools of 2026.
Gemini 3.1 Pro posted the highest GPQA Diamond and ARC-AGI-2 scores in Google's comparison table, and its 1M context and free tier make it useful for researchers and students processing large amounts of information.
Recommendation: start with Claude Pro at $20/mo if you code, add Gemini's free tier as a research companion, and add ChatGPT for async automation or enterprise features. For other AI alternatives, check our guide to best character AI alternatives or see how Grok stacks up against ChatGPT.
ChatGPT creates and edits images with OpenAI's GPT Image models, which replaced DALL-E 3. This guide covers prompting, conversational editing, sizes, pricing and alternatives.
Claude Code is Anthropic's agentic coding tool for the terminal, IDE, desktop and web. This review covers its features, how it differs from GitHub Copilot and Cursor, and pricing as of September 2026.
Claude Opus 4.8 costs $5/$25 per million tokens through the API. Kimi K2.6 and MiniMax M3 cost far less per token and each scored 10/10 on a blind debugging test on a 200,000-line codebase.
Join builders getting source-cited AI tool analysis. Never sponsored, always attributed.
Source-cited reviews and comparisons of AI tools, published by Rai Ansar.