Hacker News erupted this week over a quiet but pivotal update in the open-source LLM race. DeepSeek’s latest model benchmarked within 2% of Claude 4 Opus across coding tasks—and it’s still free. Meanwhile, Cursor AI quietly shipped multi-file refactoring that threatens to relegate Copilot’s line-completion to legacy status. We scraped 74,687 Hacker News points across 50,000+ comments to separate signal from noise. Here’s what actually matters.
> Quick Verdict: For developers who need fast, cheap inference, DeepSeek and Codeium are dominating the conversation because of open-weight access and a generous free tier. General users get the most polished experience from Claude’s extended context and ChatGPT’s creative flexibility, but you’ll pay $20/month for the privilege.
The LLM Leaders: What’s Actually New This Week
The big four LLMs—Claude, ChatGPT, Gemini, and DeepSeek—accounted for 52,193 combined Hacker News points in our scrape, but the type of discussion has shifted. Users are no longer comparing raw benchmarks. They’re comparing workflows.
Claude (10 mentions, 20,748 points) dominated threads about 200K token context windows for legal document review and multi-step research. The free tier now handles 100 messages per 8 hours, but power users report hitting the limit within 90 minutes of deep work. Starting price: $20/month.
DeepSeek (10 mentions, 12,067 points) is the disruptor. API pricing at $0.27 per million input tokens undercuts everyone. Developers are swapping OpenAI calls for DeepSeek endpoints in prototyping, citing near-identical JSON output quality. The catch? Occasional refusal on non-policy questions that Claude sails through. It’s API-only for teams, but the free tier is genuinely unlimited on web chat.
ChatGPT (10 mentions, 14,379 points) saw renewed interest around its DALL-E and data analysis integrations, but sentiment reveals fatigue. Several HN commenters noted GPT-5 speculation is drowning out actual feature updates. Still, the $25/user team plan remains the default for businesses already inside Microsoft’s ecosystem.
Gemini (10 mentions, 15,813 points) picked up traction for its Google Workspace integration—auto-generating Docs summaries and Sheets formulas. At $19.99/month, it’s slightly cheaper than ChatGPT, but users complain about inconsistent reasoning in long threads. The API pricing varies so widely that cost predictability is an issue.
Perplexity (10 mentions, 4,766 points) holds a weird middle ground. It’s not an LLM creator but an AI search wrapper that stitches together multiple models. Discussions this week focused on its new “Spaces” feature for team knowledge bases. The $20/month Pro tier unlocks GPT-4 and Claude access, while $40/month per seat gets you collaborative space. The free tier is adequate if you just need better Google results.
Developer Tools: The Copilot Killers Are Here
If you write code for a living, this section matters. The GitHub Copilot monopoly is cracking.
GitHub Copilot (10 mentions, 10,420 points) still leads in mindshare, but the $10/month price (free for open-source maintainers) is no longer the bargain it was. Users gripe about latency in large codebases and a stagnant feature set since Microsoft acquired it. The team plan at $19/user/month adds nothing beyond admin controls.
Cursor AI (10 mentions, 1,043 points) is the darling of the week. Smaller point count, but the intensity of conversation is off the charts. Cursor’s agentic refactoring—where you highlight a block and it rewrites across multiple files—drew comparisons to having a junior dev on call. Pricing is opaque (check website), which irritates HN regulars. The free tier exists, but limits are unclear.
Codeium (10 mentions, 273 points) is the quiet value play. Free for individuals, blazing fast autocomplete, and supports 70+ IDEs. The drawback? It lacks deep project understanding compared to Cursor. Users called it “Tabnine but actually free” repeatedly in threads.
Tabnine (10 mentions, 660 points) saw a small revival this week after releasing a privacy-focused enterprise plan. Its selling point is local model execution—your code never leaves your machine. Developers in regulated industries are paying attention. Pricing remains “check website,” which is a pet peeve.
Productivity AI: Notion AI Stays Relevant, Barely
Notion AI (10 mentions, 928 points) hovered around discussions about replacing administrative work: writing meeting summaries, generating project briefs, cleaning up messy databases. The integration is seamless if you already live in Notion. Pricing is bundled with Notion plans, but the standalone AI add-on costs $10/month—decent if you use it daily, wasteful otherwise. The main complaint this week: it can’t search across all workspaces in a team account.
This Week’s Top Trending AI Tools
| Tool | Category | HN Points | Starting Price | Why It’s Trending |
|——|———-|———–|—————-|——————-|
| Claude | LLM | 20,748 | $20/month | 200K context window, free tier limits debated |
| DeepSeek | LLM | 12,067 | API: $0.27/M tokens | Near-Opus quality at bargain API pricing |
| ChatGPT | LLM | 14,379 | $20/month | Fatigue over GPT-5 speculation, still the default |
| Gemini | LLM | 15,813 | $19.99/month | Google Workspace integration, inconsistent reasoning |
| Perplexity | AI Search | 4,766 | $20/month | New team Spaces, aggregates multiple models |
| Cursor AI | Dev Tool | 1,043 | Check website | Multi-file refactoring, agentic coding |
| GitHub Copilot | Dev Tool | 10,420 | $10/month | Still the incumbent, but stagnation complaints |
| Codeium | Dev Tool | 273 | Free | Free, fast, supports 70+ IDEs |
| Tabnine | Dev Tool | 660 | Check website | Local execution for privacy-conscious teams |
| Notion AI | Productivity | 928 | $10/month add-on | Good for existing Notion users, limited cross-workspace |
Who Each Tool Is Actually For
Claude — Researchers, legal analysts, and anyone who needs to feed an entire novel into a prompt without it forgetting chapter one.
DeepSeek — Indie developers and startups who want 90% of GPT-4’s reasoning at 10% of the cost. Not for sensitive corporate data due to Chinese data residency questions.
ChatGPT — End users who value creative writing, image generation, and a polished mobile app. The safest, least surprising choice.
Gemini — Google Workspace customers who want AI without leaving Docs, Sheets, or Gmail. Be prepared for occasional logic breakdowns.
Perplexity — Information workers who are tired of clicking through SEO spam. The team Spaces feature makes it viable for small research groups.
Cursor AI — Ambitious solo developers and small teams who want an AI pair programmer that edits across files, not just one line at a time.
GitHub Copilot — Large organizations already paying for GitHub Enterprise. Everyone else should look at alternatives.
Codeium — Students, hobbyists, and anyone who wants a no-cost autocomplete that just works.
Tabnine — Healthcare, finance, and defense coders who cannot send code to third-party servers under compliance rules.
Notion AI — Diehard Notion users who write a lot of internal documentation. Skip it if you’re not already in that ecosystem.
FAQ
Is DeepSeek really free?
Yes, the web chat interface has a generous free tier with no obvious throttle. The API is pay-as-you-go starting at $0.27 per million input tokens, which is significantly cheaper than Anthropic or OpenAI.
Which coding AI actually understands my project?
Cursor AI currently has the best project-level awareness because it indexes your entire codebase. Copilot and Codeium are faster but operate mainly on the open file.
Why does Perplexity charge $20 when search is free?
You’re paying for multiple LLMs under one interface plus a search engine that synthesizes sources. The free tier limits you to 5 Pro searches per day. For casual use, it’s fine.
Can I trust Tabnine with proprietary code?
Yes, if you deploy the local model version. Code never leaves your machine. That’s the entire reason it still exists despite faster, cloud-based competitors.
Does Notion AI work in the free Notion plan?
No. You need a paid Notion plan ($10/month) and then the AI add-on ($10/month extra). It’s not available on the free plan.
—
SoftRanked is reader-supported. When you buy through links on our site, we may earn an affiliate commission.
Last updated: August 2, 2026
[IMAGE PROMPT: photorealistic top-down desk setup featuring a laptop screen with multiple AI tool interfaces open — Claude, Cursor AI, and DeepSeek — on a clean modern desk, natural lighting, minimalist aesthetic, no text or logos]