LLM Usage Limits 2026: ChatGPT vs. Claude vs. Gemini (Full Comparison)

Last Updated: March 17, 2026

Every major AI platform looks different than it did three weeks ago. OpenAI dropped GPT-5.4—a single model that does what three used to. Claude’s million-token context window left beta. Google shipped its cheapest model yet. And the graveyard of retired models keeps growing: GPT-5.1 is gone, GPT-4o is gone, Claude Opus 4/4.1 is gone, and GPT-5.2 Thinking has a June expiration date.

This guide breaks down current pricing, usage limits, models, and features across ChatGPT, Claude, Gemini, Grok, and Perplexity as of mid-March 2026. It’s not a “best AI” ranking—it’s a reference for picking the platform that fits how you actually work.

What’s Changed Since February 27, 2026

March brought the biggest single model drop OpenAI has done since GPT-5 launched: GPT-5.4 fuses reasoning, coding, and computer use into one unified model. Meanwhile, Claude’s 1M context window went GA, Google shipped a new budget model, and the model graveyard keeps growing.

🧠 GPT-5.4 Released (March 5): OpenAI’s new flagship combines reasoning, coding (formerly Codex-only), and native computer-use into a single model. GPT-5.4 Thinking available on Plus/Team/Pro. GPT-5.4 Pro for Pro/Enterprise. 1M token context window. The biggest shakeup to ChatGPT’s model lineup since GPT-5.

🗑️ GPT-5.1 Retired (March 11): GPT-5.1 Instant, Thinking, and Pro all removed from ChatGPT. Conversations auto-migrate to GPT-5.3/5.4. GPT-5.2 Thinking sunsets June 5—start planning now.

📊 ChatGPT for Excel (Beta, March 5): AI embedded directly in Excel for building, analyzing, and updating spreadsheets. Google Sheets support coming soon. Business/Enterprise users can also draft emails and schedule meetings through Google and Microsoft app integrations.

🔑 Claude 1M Context Now GA: The million-token context window for Opus 4.6 and Sonnet 4.6 is now generally available at standard pricing—no beta header required. Media limit raised from 100 to 600 images or PDF pages per request.

🔌 Claude Plugin Marketplace + Memory for All: New plugin marketplace with admin controls for Team and Enterprise. Memory from chat history now available on every tier, including free.

⚡ Gemini 3.1 Flash-Lite (March 3): Google’s cheapest model yet at $0.25 per million input tokens. 45% faster than 2.5 Flash. Built for high-volume tasks where you don’t need frontier reasoning.

🎬 Grok Imagine API + 4.2 Beta: xAI launched a unified video and audio generation API for developers. Grok 4.2 entered public beta.

🖥️ Perplexity Adds GPT-5.4, Sonnet 4.6, Gemini 3.1 Pro: Perplexity Computer gained Skills, Voice Mode, and a GPT-5.3-Codex coding subagent. Pro and Max users can now access the latest models from every major lab.

📈 Claude March Usage Promotion: Anthropic doubled off-peak usage limits across all plans from March 13–27. A temporary thank-you, but a sign of how providers are experimenting with dynamic capacity.

Comparison Table

🔑 Key Insights from the Mid-March 2026 Comparison

• GPT-5.4 is a consolidation play. Reasoning + coding + computer use in one model. OpenAI is collapsing its multi-model lineup into fewer, more capable systems. The era of choosing between “the thinking model” and “the coding model” may be ending.
• The 1M context club has three members. GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6 (now GA) all offer 1M tokens. Grok leads at 2M. Context window size is becoming table stakes at the $20 tier.
• Legacy models keep disappearing. GPT-5.1 gone (March 11), GPT-4o gone (Feb 13), Claude Opus 4/4.1 removed, Gemini 3 Pro deprecated. GPT-5.2 Thinking sunsets June 5. If you’re relying on older model behaviors, the clock is ticking.
• Spreadsheet and productivity integration is the new battleground. ChatGPT for Excel, Claude in Excel/PowerPoint with cross-app context, Gemini in Workspace with AI-assisted drafting—all three platforms now embed directly in office tools. This is where AI stops being a chat window and starts being a work tool.
• Perplexity’s multi-model strategy widens its model gap. With GPT-5.4, Sonnet 4.6, Gemini 3.1 Pro, and Opus 4.6 all available on one platform, Perplexity Pro at $20/mo gives you access to more frontier models than any single-provider plan.
• The $200+ tier remains “unlimited.” ChatGPT Pro ($200), Claude Max 20x ($200), SuperGrok Heavy ($300), and Perplexity Max ($200) all remove meaningful usage caps for power users.

At a Glance: The $20 Tier (Plus a Budget Option)

Before diving into each platform, here’s how the mid-tier subscriptions stack up—plus a budget option.

💡 Budget Option: ChatGPT Go ($8/mo)

If $20/month feels steep, ChatGPT Go offers GPT-5.2 Instant at $8/month with 10x more messages than free. You won’t get GPT-5.4, reasoning mode, or Codex (those require Plus), but for everyday tasks, it’s solid value. Be aware: ads will appear in this tier. And with GPT-5.4 widening the capability gap between Go and Plus, the $12 difference buys you a lot more than it used to.

$20/Month AI Showdown

Comparing mid-tier AI subscriptions • Mid-March 2026

Platform Breakdown: ChatGPT (OpenAI)

March was transformative. GPT-5.4 consolidates everything OpenAI has shipped over the past six months—reasoning, Codex-level coding, and native computer use—into one flagship model. Then they retired GPT-5.1, pushed ChatGPT for Excel into beta, and started the clock on GPT-5.2 Thinking’s June sunset. If you haven’t updated your workflow since February, it looks completely different now.

ChatGPT Free

Access to GPT-5.2 Instant with roughly 10 messages every 5 hours. Basic DALL-E image generation and web search included. When you hit limits, you’re downgraded to a lighter model automatically. Ads are showing for logged-in US users. Good for casual use, but you’ll hit walls quickly.

ChatGPT Go ($8/month)

OpenAI’s budget tier. You get:

• GPT-5.2 Instant (the fast variant—no Thinking or Pro modes)
• 10x more messages, file uploads, and image creation than free
• Longer memory and context window
• No reasoning mode, no Codex access, no GPT-5.4
• Ads will appear in this tier

At $8/month, it’s the cheapest way into serious GPT territory—but the gap between Go and Plus just got much wider with GPT-5.4.

ChatGPT Plus ($20/month)

The sweet spot, and it just got a major upgrade. GPT-5.4 Thinking is now the default reasoning model—a unified frontier model that combines advanced reasoning, coding, and the ability to autonomously operate computers and software. 1M token context window. Roughly 150 messages per rolling 3-hour window. Full DALL-E and GPT Image 1.5 access. Includes Codex with GPT-5.3-Codex and the new ChatGPT for Excel beta. Ad-free.

The rolling window system means your quota constantly refreshes. Send 50 messages at 9am, and those slots open back up at noon. GPT-5.4 Thinking replaces the need for separate reasoning and coding models—it handles both natively.

Important: GPT-5.2 Thinking will be fully retired on June 5, 2026. All users will be migrated to GPT-5.4. If you have workflows that rely on GPT-5.2-specific behavior, plan your migration now.

ChatGPT Pro ($200/month)

Unlimited GPT-5.4 usage plus GPT-5.4 Pro for maximum performance on complex tasks. Includes o3 Pro Mode for demanding reasoning, GPT-5.3-Codex-Spark on Cerebras hardware at 1000+ tokens/second, and native computer use for autonomous software operation. Priority access during peak times. Ad-free forever.

What’s new for ChatGPT (March 2026):

• GPT-5.4 (March 5): New unified frontier model. Combines reasoning, coding (absorbing GPT-5.3-Codex capabilities), and native computer use. 1M token context. Available as GPT-5.4 Thinking (Plus/Team/Pro) and GPT-5.4 Pro (Pro/Enterprise). OpenAI calls it their most capable model for professional work.
• GPT-5.3 Instant (March 3): Updated conversational model with improved follow-up tone. Reduced teaser-style phrasing in responses.
• GPT-5.1 Retired (March 11): GPT-5.1 Instant, Thinking, and Pro removed from ChatGPT. Existing conversations auto-migrate to GPT-5.3/5.4 equivalents.
• ChatGPT for Excel (Beta, March 5): AI embedded directly in Excel for building, analyzing, and updating spreadsheets. Limited to US, Canada, Australia during beta. Google Sheets support coming soon.
• Google/Microsoft Write Actions: Business and Enterprise users can now draft emails, create docs and spreadsheets, and schedule meetings via Google and Microsoft apps directly from ChatGPT. Disabled by default—admins enable in Workspace settings.
• Interactive Visual Modules: ChatGPT can now present interactive visuals for math and science topics. 70+ topics at launch. Rolling out to all logged-in users.
• GPT-5.2 Thinking Sunset: Scheduled for June 5, 2026. Plan accordingly if workflows depend on 5.2-specific behavior.

Platform Breakdown: Claude (Anthropic)

March’s headline for Claude: the 1M token context window is no longer beta. It’s GA for Opus 4.6 and Sonnet 4.6 at standard pricing, no special headers required. Add in a plugin marketplace, memory for free users, and Excel/PowerPoint add-ins that now share full cross-app context—and Claude’s ecosystem keeps expanding beyond just “the coding model.”

Claude Free

Sonnet 4.6 access with roughly 15-45 messages every 5 hours depending on conversation complexity and system load. Off-peak hours give you more; business hours can drop to the lower end. Memory from chat history now available on free tier—a meaningful quality-of-life upgrade. No web search. Good for trying the model, but limits feel restrictive for any serious work.

Claude Pro ($20/month)

Around 45 messages per 5-hour window, plus access to Extended Thinking for complex reasoning. Weekly caps sit on top of the 5-hour limits. Includes Opus 4.6—Anthropic’s most capable model—now with a 1M token context window at standard pricing (no longer beta).

Also includes:

• Cowork for agentic desktop tasks (macOS only)
• Claude Code for terminal-based coding
• Plugin marketplace for extending Claude’s capabilities
• Google Workspace integration (Docs, Gmail, Drive)
• Extra usage option—continue at API rates when you hit limits

Claude Max ($100/month and $200/month)

The $100 tier (5×) and $200 tier (20×) both include full Opus 4.6 access with:

• 1M token context window (GA)—now standard, no beta qualifier
• Agent teams (research preview)—split tasks across multiple agents working in parallel
• Adaptive thinking and effort controls
• Cowork with full functionality and higher task output limits
• Priority access during peak traffic
• Extra usage option at API rates
• Up to 600 images or PDF pages per request with 1M context

Best for developers doing extensive code work or researchers handling massive documents. The 1M context on Opus makes it genuinely useful for entire codebases—and it’s no longer gated behind a beta flag.

What’s new for Claude (March 2026):

• 1M Context Window Now GA: Generally available for Opus 4.6 and Sonnet 4.6 at standard pricing. No beta header required. Dedicated 1M rate limits removed—your standard account limits now apply across every context length. Media limit raised from 100 to 600 images or PDF pages per request.
• Memory for All Users: Memory from chat history is now available on every tier, including free. Import and export your memory data.
• Plugin Marketplace: New marketplace plus admin controls for Team and Enterprise plans.
• Claude in Excel & PowerPoint Updated: Add-ins now share full conversation context across applications. Support for LLM gateway connections via Amazon Bedrock, Google Vertex AI, and Microsoft Foundry.
• Opus 4/4.1 Removed: Deprecated from the model selector and Claude Code.
• Sonnet 4.6 Upgraded: Described as the “most capable Sonnet yet” with full upgrades across coding, computer use, long-context reasoning, agent planning, and knowledge work.
• Self-Serve Enterprise: Any organization can now purchase Enterprise directly on the website—no sales conversation required.
• Inline Visualizations: Claude can now create custom charts, diagrams, and visualizations directly in responses.
• March Usage Promotion (March 13–27): Temporary doubling of off-peak usage across Free, Pro, Max, and Team plans.
• Claude Partner Network: $100M commitment for training, certification, and go-to-market support for enterprise partners.

Platform Breakdown: Gemini (Google)

Gemini’s March was more about filling out the lineup than dropping bombshells. Gemini 3.1 Flash-Lite gives developers a genuine budget model, Gemini in Workspace got a significant upgrade, and Chrome integration expanded globally. The big model news—3.1 Pro—landed in late February and is now the default across Pro and Ultra tiers.

Gemini Free

Access to Gemini 3 Flash—Google’s frontier model optimized for speed. PhD-level reasoning at Flash-level latency. Free users in the US also get limited access to Gemini 3 Pro (“Thinking” mode) and Nano Banana Pro image generation with daily limits.

Developer bonus: Free tier users get access to Google Antigravity (agentic IDE) with weekly rate limit refreshes.

Google AI Pro ($19.99/month)

Full Gemini 3.1 Pro access—Google’s most capable Pro-tier model with over double the reasoning scores of its predecessor. 100 prompts/day, 1M token context window, Nano Banana Pro (100 images/day), Veo 3.1 Fast video generation (3 videos/day), 20 Deep Research reports/day. Gemini in Chrome for AI-assisted browsing, now available in Canada, New Zealand, India, and 50+ languages.

Gemini in Workspace now lets you draft documents with context from your files and emails, match writing style across documents, and create spreadsheets and presentations with AI assistance. Available for Pro and Ultra.

Includes 2TB Google One storage, Google Home Premium Standard, and priority Antigravity access with 5-hour refresh cycles.

Google AI Ultra ($124.99/3 months, ~$42/month)

Premium tier with highest limits. Gemini 3 Deep Think mode for science, research, and engineering. Gemini 3.1 Pro with the highest usage limits. Veo 3.1 for video generation with sound. 30TB storage, YouTube Premium, Agent Mode, highest Antigravity rate limits, and early access to new features.

Regular price is $249.99/month—the $124.99/3 months is an introductory offer.

What’s new for Gemini (March 2026):

• Gemini 3.1 Flash-Lite (March 3): Budget model at $0.25/1M input tokens and $1.50/1M output. 45% faster than 2.5 Flash with a 2.5x faster time to first token. 1M token context, 64K output. Designed for high-volume tasks like translation, content moderation, and UI generation.
• Gemini 3 Pro Preview Shut Down (March 9): Deprecated as planned. The -latest alias now points to Gemini 3.1 Pro Preview. If you hadn’t migrated, you’ve been auto-migrated.
• Gemini in Workspace Major Update: New capabilities in Docs, Sheets, Slides, and Drive. Source-aware drafting (pull from files, emails, and web), style matching, format matching, and AI-assisted spreadsheet creation. Pro and Ultra subscribers.
• Gemini in Chrome Expanded: Now available in Canada, New Zealand, and India with 50+ additional languages. Auto browse feature helps automate tasks while keeping you in control.
• Computer Use Support: Now available in Gemini 3.1 Pro Preview and Gemini 3 Flash Preview via the developer API.
• Student Offer Continues: Free AI Pro for eligible students through July 2026 in Indonesia, Japan, UK, and Brazil.

Platform Breakdown: Grok (xAI / SpaceX)

March was quieter than February for Grok—which, after a SpaceX acquisition and a new rapid-learning model, is understandable. The main developments: Grok 4.2 entered public beta, and the Imagine API launched for developers looking to build with Grok’s video and audio generation capabilities.

Grok Free

Grok 3 with roughly 10 requests every 2 hours. Real-time X integration gives it unique access to trending conversations. Aurora image generation with limits. Best for X power users who want AI integrated into their social feed.

SuperGrok ($30/month or $300/year)

Grok 4.20 Beta alongside the new Grok 4.2 Beta. The 4.20 Beta features a “rapid learning” architecture that improves weekly from user feedback. The 4.2 Beta is the newer public beta with further refinements. Includes “Thinking” mode (1483 Elo on LMArena), higher usage limits, 256K context window, DeepSearch, advanced reasoning, and the Imagine model for image/video generation.

Note: Image generation of real people remains restricted following the January controversy. Safety investigations continue in 7+ countries.

SuperGrok Heavy ($300/month)

Grok 4.1 Heavy with the massive 2M token context window—the largest on any consumer platform. Unlimited usage, multi-agent capabilities, maximum compute priority, extended thinking. Early access to features including Grok 5 when it launches (currently in training on Colossus with 1M+ H100 GPU equivalents).

What’s new for Grok (March 2026):

• Grok 4.2 Beta: Now in public beta. Check xAI documentation for the latest updates on capabilities vs. Grok 4.20 Beta.
• Grok Imagine API: Unified video and audio generation suite for developers. Image-to-video, text-to-video, restyling, object editing, and motion control. Optimized for latency, concurrency, and cost.
• Grok 5 Still in Training: Confirmed training on Colossus I and II. No release date announced.
• Safety Investigations Ongoing: Regulatory probes continue in 7+ countries following the December 2025–January 2026 image generation controversies. Image generation of real people remains restricted.

Platform Breakdown: Perplexity

Perplexity keeps adding models from every major lab. March brought GPT-5.4, Claude Sonnet 4.6, and Gemini 3.1 Pro to Pro and Max subscribers. Computer—the 19-model autonomous agent—got Skills, Voice Mode, and a coding subagent. The multi-model strategy is becoming Perplexity’s clearest differentiator.

Perplexity Free

Unlimited basic searches with Sonar. 5 Pro searches per day using advanced models. Good for quick research, but you’ll burn through Pro searches fast on complex topics.

Perplexity Pro ($20/month)

300+ Pro searches daily with your choice of GPT-5.4, Claude Sonnet 4.6, Gemini 3.1 Pro, and other frontier models. Deep Research mode for comprehensive analysis. Unlimited file uploads. $5 monthly API credit. Includes email assistant, patent research access, and Seedream 4.5 image generation.

Perplexity Max ($200/month)

Unlimited everything. Access to o3-pro, Opus 4.6, GPT-5.4. Unlimited Labs for dashboards, spreadsheets, and presentations. Sora 2 Pro video generation. Comet browser agent defaults to Opus 4.6. Perplexity Computer with 19-model orchestration, Skills, Model Council with memory, Voice Mode, and GPT-5.3-Codex coding subagent. 10,000 monthly credits plus a one-time 20,000 credit bonus.

What’s new for Perplexity (March 2026):

• GPT-5.4 Added: Available for Pro and Max subscribers alongside existing multi-model roster.
• Claude Sonnet 4.6 and Gemini 3.1 Pro Added: Expanding the model selection further.
• Perplexity Computer Updates: Skills for reusable workflows, Model Council with memory across sessions, Voice Mode for hands-free interaction, and GPT-5.3-Codex as a coding subagent.
• Comet for Android Expanded: Voice chat across tabs, cross-tab summarization, AI assistant integration.
• Deep Research on Opus 4.6: Continues to run on Claude Opus 4.6 with state-of-the-art benchmark performance.

Understanding New Usage Paradigms

AI usage limits in 2026 aren’t just about “messages per hour” anymore. Here’s what’s actually happening under the hood.

Rolling Windows vs. Fixed Resets

ChatGPT uses a rolling 3-hour window for Plus users. Each message “expires” from your quota exactly 3 hours after you sent it. This means your available capacity fluctuates constantly—send 50 messages between 9-10am, and those slots free up between noon and 1pm.

Claude uses a 5-hour session window for Pro users, with additional weekly caps on top. Max users have weekly limits that reset every 7 days. Anthropic also offers “extra usage”—when you hit your limit, you can continue at standard API rates instead of waiting. The March usage promotion temporarily doubled off-peak limits, hinting at future dynamic pricing models.

Gemini uses daily prompt limits (100/day for Pro) with no rolling mechanism—simpler to understand but less flexible during intensive work sessions.

Context Windows: When Size Matters

The context window race reached a convergence point in March. Three platforms now offer 1M tokens at the $20 tier:

• Grok SuperGrok Heavy: 2M tokens — largest consumer context window available
• GPT-5.4: 1M tokens — massive jump from GPT-5.2’s 256K
• Gemini 3.1 Pro: 1M tokens — massive and well-optimized
• Claude Opus 4.6: 1M tokens (GA) — no longer beta, standard pricing
• Claude Sonnet 4.5/4.6: 200K-1M tokens — 1M available via API
• ChatGPT (GPT-5.2 standard): 128K tokens — still available but being sunsetted

Longer context means you can process entire codebases, books, or research paper collections in a single conversation. But longer context also costs more compute—which is why models like Anthropic charge 2x for inputs over 200K tokens on the API.

The shift from 256K to 1M on GPT-5.4 is the biggest context upgrade any platform has made in a single release. It means Plus subscribers can now handle roughly the same document sizes as Gemini and Claude users—a gap that existed for months is now closed.

Agentic AI: The New Frontier

Every major platform now has autonomous task execution:

• Claude Code + Agent Teams: Terminal-based coding agent that can spawn sub-agents working in parallel. Opus 4.6 holds the METR record for longest autonomous task horizon at 14.5 hours.
• OpenAI Codex (GPT-5.3) + GPT-5.4: Agentic coding in the Codex app, CLI, IDE extension, and macOS desktop app. GPT-5.4 adds native computer use—the model can autonomously navigate software and browsers.
• Google Antigravity: Agent-first IDE where AI plans, executes, and verifies development workflows. Free tier with weekly limits.
• Grok Agent Tools: API-level agent capabilities with real-time X data access. Grok 4.20 Beta adds 4-agent parallel collaboration. Grok Imagine API extends agentic capabilities to video and audio.
• Perplexity Labs + Computer: Autonomous research, dashboard creation, and report generation. Computer takes it further with 19-model orchestration, subagent creation, Skills, and Voice Mode.

These tools are compute-intensive. Using Cowork, Codex, or Perplexity Computer burns through your quota faster than regular chat. If you rely heavily on agentic features, the Max/Pro tiers start making financial sense.

Office Integration: AI Moves Into Your Spreadsheets

This is the quieter trend that might matter more than any model release. All three major platforms now embed directly in productivity tools:

• ChatGPT for Excel (beta): Build, analyze, and update spreadsheets with AI directly in Excel. Google Sheets coming soon.
• Claude in Excel & PowerPoint: Updated add-ins share full conversation context across apps. Native Excel operations like pivot tables and conditional formatting.
• Gemini in Workspace: Source-aware drafting in Docs, AI-assisted spreadsheets in Sheets, style matching, and format matching.

This shifts AI from “open a separate tab” to “it’s already in my workflow.” For enterprise adoption, this is arguably more important than benchmark scores.

The Evolution Continues: What’s Next

Mid-March 2026 feels like a consolidation moment. The major labs aren’t just shipping new models—they’re simplifying. Fewer models doing more things. Bigger context windows at lower tiers. Office tools with AI baked in. Here’s what to watch.

Four Trends to Watch

1. Model consolidation is accelerating. GPT-5.4 fuses reasoning, coding, and computer use into one system. Claude Opus 4.6 handles coding, long documents, and agent tasks in a single model. The days of picking separate models for separate tasks are fading—and that makes the $20 tier genuinely powerful.

2. Office integration is the real battleground. ChatGPT for Excel, Claude in Excel and PowerPoint, Gemini in Workspace—every major platform now lives inside your productivity tools. This shifts AI from “open a chat window” to “it’s already in my spreadsheet.” Watch for adoption data here; this is where enterprise revenue will come from.

3. Context windows converged at 1M. Three platforms now offer 1M tokens at the $20 tier. Grok holds the lead at 2M. You can now process full codebases, entire research paper collections, or complete book manuscripts in a single conversation on any major platform. The question shifts from “can I fit this?” to “should I?”

4. The model graveyard is getting crowded. GPT-4o, GPT-5.1, Claude Opus 4/4.1, Gemini 3 Pro Preview—all gone in the span of weeks. GPT-5.2 Thinking sunsets June 5. If your automation or workflows reference specific model names, audit them now.

How to Stay Current

LLM usage limits change constantly. What’s accurate today may shift next week:

Monitor your usage. Most platforms now show real-time meters. Check them weekly to understand your patterns before you hit walls.

Follow changelogs. Each platform has release notes. Bookmark them. Changes drop without warning.

Test before committing. Free tiers let you try frontier models risk-free. Use them to validate fit before paying.

Official sources:

• OpenAI: https://help.openai.com/
• Anthropic: https://docs.anthropic.com/
• Google: https://ai.google.dev/
• xAI: https://docs.x.ai/
• Perplexity: https://www.perplexity.ai/

Final Thoughts

Picking the right AI platform in mid-March 2026 comes down to what you actually do with it every day.

If you want maximum value under $20: ChatGPT Plus just leapfrogged with GPT-5.4 Thinking—1M context, unified reasoning and coding, and computer use at $20/mo. Claude Pro gives you Opus 4.6 with 1M context (GA) plus Cowork and Claude Code. Google AI Pro includes Gemini 3.1 Pro with deep Workspace integration. All three are stronger than they were a month ago.

If coding is your primary use case: Claude with Opus 4.6 and agent teams still leads—METR’s 14.5-hour autonomous task horizon record stands. GPT-5.4 is a strong challenger with integrated coding capabilities. Google Antigravity rounds out options for the Google ecosystem.

If you work with massive documents: Grok’s 2M token context (SuperGrok Heavy) remains unmatched. GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6 all offer 1M—a three-way tie at the premium tier.

If research is your priority: Perplexity’s multi-model approach—GPT-5.4, Opus 4.6, Sonnet 4.6, Gemini 3.1 Pro all in one place—makes it the most versatile research platform. Deep Research on Opus 4.6 plus Perplexity Computer for autonomous workflows.

If you’re in Google’s ecosystem: Gemini in Workspace now drafts documents from your files and emails, matches your writing style, and creates spreadsheets with context. That’s not a chat feature—that’s a productivity tool. The 3.1 Pro model makes it genuinely competitive on reasoning too.

If you need your AI in your spreadsheets: ChatGPT for Excel (beta), Claude in Excel with native operations, and Gemini in Sheets all shipped or updated recently. This is the most competitive office-integration moment we’ve seen.

If ads bother you: ChatGPT Plus ($20+), Claude (all tiers), Gemini (all tiers), Grok (all tiers), and Perplexity (all tiers) remain ad-free. Only ChatGPT Free and Go show ads. Perplexity notably dropped ads entirely to protect answer integrity.

The landscape will look different in April—especially with GPT-5.2 Thinking’s June sunset approaching and Grok 5 in training. Use this guide as a mid-March 2026 snapshot, verify with official sources before committing, and don’t be afraid to switch when your needs change.

Last Updated: March 17, 2026

Note: Usage limits and features change frequently. Always verify current information with official documentation from OpenAI, Anthropic, Google DeepMind, xAI, and Perplexity.

Change Log (March 17, 2026 Update)

• Updated ChatGPT primary model from GPT-5.2 to GPT-5.4 across Plus and Pro tiers. Context window now 1M tokens.
• Documented GPT-5.1 retirement (March 11) and GPT-5.2 Thinking sunset date (June 5, 2026).
• Added ChatGPT for Excel beta and Google/Microsoft write actions for Business/Enterprise.
• Updated Claude 1M context from beta to GA for Opus 4.6 and Sonnet 4.6. Media limit now 600.
• Added Claude plugin marketplace, memory for free users, inline visualizations, and self-serve Enterprise.
• Added Gemini 3.1 Flash-Lite ($0.25/1M input) and Workspace integration upgrades.
• Confirmed Gemini 3 Pro Preview shut down (March 9).
• Added Grok 4.2 Beta and Grok Imagine API.
• Updated Perplexity model access: GPT-5.4, Sonnet 4.6, Gemini 3.1 Pro added. Computer gained Skills and Voice Mode.
• Refreshed comparison table, context window rankings, and all NEW badges for March.

Sources

All changes verified through:

• OpenAI: GPT-5.4 Announcement, ChatGPT Release Notes, ChatGPT Pricing
• Fortune: OpenAI launches GPT-5.4
• Anthropic: Claude Release Notes, API Changelog, Models Overview
• Google: Gemini 3.1 Flash-Lite, Gemini 3.1 Pro, Workspace Updates, API Release Notes
• xAI: Release Notes, Models and Pricing, xAI News
• Perplexity: Changelog, Enterprise Pricing
• TechCrunch: Perplexity Computer
• Wikipedia: Claude (language model), Gemini (language model)