Perplexity just dropped a stack of updates on February 13 that make its AI feel less like a forgetful mate and more like that colleague who nails every detail from last week’s meeting. We’re talking Opus 4.6 powering Deep Research, memories hitting 95% recall accuracy, and new models like GPT-5.2 for coders grinding SWE-Bench tasks. If you’re knee-deep in campaign briefs or GitHub audits, these changes mean less faffing about and more actual results, honestly, it’s the kind of upgrade that saves you from rewriting the same prompt for the third time.
I remember last Tuesday, buried in a client report on Nvidia stock trends, asking Perplexity to pull threads from a chat two weeks back. Pre-update, it ghosted half the context; now, it spits back specifics like the exact P/E ratio we debated over flat whites. Too real? That one thread where it recalled my offhand note about Shopify inventory syncs, saved me 20 minutes of scrolling history.
—
✅ Opus 4.6 Deep Research Upgrade
Deep Research now runs on Opus 4.6, smashing benchmarks like Google DeepMind Deep Search QA and Scale AI Research Rubric for top accuracy. It pairs cutting-edge models with Perplexity’s search engine and sandbox setup. Max users get it now, Pro rolling out soon.[1][3]
Developers can dissect GitHub commit histories for patterns without manual digging. Marketers generate campaign briefs by auto-summarising competitor onboarding flows. Analysts walk through web dashboards to spot data trends, like Nvidia stock shifts. Researchers tackle marathon training plans pulling your past fitness chats.
—
✅ Improved Memory Engine
Perplexity builds richer memories and recalls them in 95% of cases, up from 77%, while making half as many to cut clutter. Rolled out to Max and Pro users.[1][2]
Content writers ask for book recs based on last month’s reads it actually remembers. Developers reference prior code snippets without re-explaining the stack. Marketers sync past campaign notes for fresh briefs. Here’s the key stats:
- Recall accuracy: 95% (was 77%)[1][2]
- Memories created: 50% fewer[1][2]
- Better at time understanding and approximations[2]
—
✅ Memory in Model Council
Model Council now taps personal memories for each model, synthesising sharper answers from multiple AIs. Great for decisions needing context.[3]
Analysts query retirement portfolios with your risk tolerance from prior chats. Developers evaluate MBA programs factoring old career notes. Marketers brainstorm strategies recalling last quarter’s Shopify sync fails. Try prompts like ‘top MBA programs for me?’, it pulls your specifics.[3]
—
✅ GPT-5.2 and Seedream 4.5 Models
New options for Pro and Max: GPT-5.2 crushes coding on SWE-Bench Pro and multi-step reasoning; Seedream 4.5 from Bytedance delivers cinematic images with better instruction follow.[1]
Developers debug complex software eng tasks. Content writers craft punchy visuals for blog headers, like a data-flavoured graph for inventory reports. Marketers generate campaign imagery that nails spatial layouts, no more wonky proportions.
—
✅ Kimi K2.5 Open-Source Reasoning
Fresh state-of-the-art model from Moonshot AI, hosted on Perplexity’s US stack for low latency and security. Rolls to Max then Pro.[1][3]
Researchers handle long-form analysis with reliable speed. Developers test reasoning on call transcript summaries. Everyone benefits from tighter control over speed for real tasks like generating Shopify reports.
—
✅ Comet Browser Agent on Opus 4.6
Max users default to Opus 4.6 for Comet, boosting reasoning for web tasks. Pick Sonnet 4.5 too.[2]
Analysts probe dashboards for patterns. Developers audit GitHub histories. Marketers assess competitor flows, spotting friction in onboarding like that clunky HubSpot demo last week.
—
✅ Improved History Search and Android Stock Graphs
Search past threads smarter, with previews jumping to exact spots. Android app shows revamped Nvidia stock charts inline.[3]
We’ve all been there, hunting that one thread. Now it’s instant. Traders on Android get expanded stock info without app-switching.
—
Dive into Perplexity at perplexity.ai, test these on your next brief or code review, and hit feedback if something clicks, or doesn’t. Subscribe for the next round; these updates move fast.
