AI Subscription Changelog: March 2026

March 2026 was a busy month for AI subscriptions. Two major model launches, a new pricing tier announcement, billing changes, and feature updates across most providers. Here is every change that matters to subscribers, organized by provider.

ChatGPT (OpenAI)

Model updates:

• GPT-5.4 received a minor update (5.4.1) improving instruction following and reducing hallucination rates on factual queries. OpenAI reported a 12% improvement on their internal accuracy benchmark. Available on Plus and Pro tiers.
• GPT-5 mini received performance improvements, notably faster response times (approximately 20% faster on average).

Feature changes:

• ChatGPT Go plan announced on March 25 with an April 1 launch date. $5/month for GPT-5 access, file uploads, and web browsing. See our full Go analysis.
• Custom GPTs now support persistent memory across conversations (Plus and Pro only)
• Canvas mode expanded to include code editing with syntax highlighting for 12 additional languages

Pricing changes:

• No changes to existing tier pricing
• Go tier ($5/month) is the first new consumer tier since Pro launched

Impact: The Go announcement is the biggest news. For existing subscribers, GPT-5.4.1 and Custom GPT memory are incremental improvements. Full details on our ChatGPT pricing page.

Claude (Anthropic)

Model updates:

• Opus 4.6 launched on March 12. This is a significant model update with major improvements in coding, analysis, and long-context tasks. Our Opus 4.6 coding test covers the technical details.
• Sonnet 4.5 received a stability update reducing error rates on complex prompts
• Haiku 4 remains unchanged

Feature changes:

• Annual billing launched for Pro ($204/year, saving 15%) and Team ($240/user/year, saving 20%) plans
• Projects feature updated with better file organization and search within project contexts
• Artifacts now support interactive code previews (React, HTML/CSS/JS)
• Max plan added higher rate limits for Opus 4.6 (approximately 2x the previous Opus 4 limits)

Pricing changes:

• Annual billing introduced (Pro: $17/month annual vs $20 monthly; Team: $20/user/month annual vs $25 monthly)
• No changes to monthly pricing for any tier

Impact: Opus 4.6 is a genuine leap for coding and professional users. Annual billing makes Claude more competitive on price. See our Claude pricing page for the complete breakdown.

Gemini (Google)

Model updates:

• Gemini 2.5 Flash received a speed update, now approximately 30% faster on short queries
• Gemini 2.5 Pro context window expanded from 1M to 1.5M tokens for Advanced subscribers
• No changes to the Ultra model (2.5 Ultra still in limited preview)

Feature changes:

• NotebookLM Plus added audio overview generation from uploaded documents
• Gemini in Google Docs now supports document formatting suggestions (not just text generation)
• Gemini Advanced subscribers get priority access to new Google AI experiments
• AI Ultra tier pricing confirmed at $249.99/month (launching April 3)

Pricing changes:

• Gemini AI Ultra confirmed at $249.99/month (the most expensive consumer AI subscription)
• Advanced pricing unchanged at $19.99/month
• Free tier unchanged

Impact: The 1.5M context window expansion for Advanced is significant for users working with large documents. Ultra pricing will be a niche purchase. Full details on our Gemini pricing page.

Perplexity

Model updates:

• Internal model improvements focused on citation accuracy (Perplexity reports 15% fewer incorrect citations)
• Pro Search now uses an updated reasoning pipeline for complex multi-step queries

Feature changes:

• Pro search limit increased from 500 to 600 per day
• Collections feature updated with better sharing and collaboration options
• Mobile app added offline access for saved searches
• API access expanded for Pro subscribers (higher rate limits)

Pricing changes:

• No price changes
• Annual plan remains at $200/year ($16.67/month)

Impact: Incremental improvements across the board. The extra 100 daily Pro searches benefits heavy researchers. No major changes to report. See our Perplexity pricing page.

Grok (xAI)

Model updates:

• Grok 3.5 launched on March 18. Notable improvement in reasoning, coding, and conversational quality. Benchmark results show it closing the gap with GPT-5 on several metrics.
• Grok 3 (previous model) phased out for Premium subscribers; Free tier still uses Grok 3

Feature changes:

• DeepSearch updated with broader web access beyond X/Twitter
• Image understanding improved (better at analyzing charts, diagrams, and screenshots)
• Grok now available as a standalone web app (previously X-only for some users)

Pricing changes:

• No price changes
• Premium+ remains $16-22/month (bundled with X Premium+)

Impact: Grok 3.5 is a meaningful upgrade that justifies the Premium+ subscription more strongly. Bundled with X Premium+ at $16-22/month, the value proposition improved without a price increase. See our Grok pricing page.

Copilot (Microsoft)

Model updates:

• Copilot updated its underlying model to GPT-5 (previously used a GPT-4 variant for some tasks)
• Copilot for coding (GitHub Copilot) updated with improved multi-file awareness

Feature changes:

• Copilot Vision launched in preview for Pro subscribers (visual understanding of screen content)
• Copilot in Excel added natural language formula creation
• Copilot in PowerPoint improved slide generation from outlines
• Teams integration expanded with meeting summary improvements

Pricing changes:

• No price changes
• Pro remains $20/month

Impact: Copilot Vision is the most interesting addition. The Microsoft 365 integrations continue to improve incrementally. Unless you use Microsoft 365 daily, these changes are not compelling enough to subscribe. See our Copilot pricing page.

DeepSeek

Model updates:

• DeepSeek V3.5 released with improved multilingual performance
• R1 reasoning model received efficiency improvements (faster responses, same quality)

Feature changes:

• Web interface updated with conversation branching (explore different response paths)
• File upload support added (previously text-only)
• API pricing adjusted slightly (lower costs for V3.5 compared to V3)

Pricing changes:

• Consumer product remains free
• API pricing reduced for the V3.5 model

Impact: DeepSeek continues to offer remarkable value at zero cost. The file upload addition brings it closer to feature parity with paid competitors. See our DeepSeek pricing page.

Mistral

Model updates:

• Mistral Large 3 launched with improved code generation and reasoning
• Mistral Medium updated for better instruction following

Feature changes:

• Le Chat Pro added document analysis (PDF and DOCX upload)
• Custom instructions feature launched (similar to ChatGPT’s custom instructions)
• Le Chat Free increased daily message limits

Pricing changes:

• No price changes
• Le Chat Pro remains $15/month

Impact: Mistral Large 3 is a solid model update. Document analysis brings Le Chat Pro closer to competitors. The European privacy focus remains Mistral’s differentiator. See our Mistral pricing page.

Meta AI

Model updates:

• Llama 4 integrated into Meta AI across all platforms (previously Llama 3.x on some surfaces)
• Improved image understanding capabilities

Feature changes:

• Meta AI expanded to Threads (previously available on Facebook, Instagram, WhatsApp, and Messenger)
• Voice interaction improved on WhatsApp
• Meta AI Studio updated for custom AI character creation

Pricing changes:

• No changes, remains free
• No paid tier announced

Impact: Llama 4 integration across all Meta platforms is significant for the hundreds of millions of users who encounter Meta AI casually. Quality improved noticeably for everyday tasks. See our Meta AI page.

Summary: What Mattered in March

The three biggest developments in March 2026:

1. Claude Opus 4.6 — The strongest coding model available to consumers. This shifts the Claude vs. ChatGPT comparison meaningfully for developers.

2. ChatGPT Go announcement — The $5 tier creates a new market segment. Budget AI subscriptions are now a real category.

3. Grok 3.5 — At $16-22/month (bundled with X Premium+), Grok is now a legitimate recommendation rather than a curiosity. The model quality caught up with the price.

For the full pricing picture across all providers, visit our pricing hub. For April changes, see our April pricing update. We publish this changelog monthly — check back in early May for the April edition.