Claude just took over the conversation. Again. With an update that nudges its already massive context window past 1 million tokens and significantly improved code generation, the community absolutely lit up—20,745 points on Hacker News says it all. But that’s not the whole story. DeepSeek quietly dropped API pricing that undercuts everyone, Perplexity made its research mode noticeably smarter, and developer tools like Cursor AI are finally worth the hype. This week wasn’t about flashy demos; it was about shipping features that actually make a difference.
> Quick Verdict: Claude is the clear winner for general-purpose power users who need huge context and nuanced reasoning. DeepSeek is the undisputed budget champion for developers building AI-native apps. ChatGPT and Gemini remain safe, multimodal workhorses, while GitHub Copilot dominates the IDE. Pick based on whether you need depth, cost-efficiency, or seamless integration.
This Week’s AI Launches at a Glance
| Tool | Starting Price | Free Tier | Best For | Why It’s Buzzing This Week |
|——|—————|———–|———-|—————————-|
| Claude | $20/mo (Pro) | Yes | Research, long-form writing, complex coding | 1M+ token context window; HN score 20,745 |
| DeepSeek | API: $0.27/M tokens | Yes (chat) | Cost-sensitive API projects, startups | New API pricing, open-weight models; 12,067 HN points |
| ChatGPT (GPT-5) | $20/mo (Plus) | Yes | Multimodal tasks, everyday productivity | Refined plugin ecosystem; 14,379 points |
| Gemini | $19.99/mo (Advanced) | Yes | Google Workspace integration, multimodal search | Deep Google Drive/Gmail hooks; 15,813 points |
| Perplexity | $20/mo (Pro) | Yes (limited) | Research, fact-checking, real-time search | Upgraded Pro search with source synthesis; 4,766 points |
| Cursor AI | Free tier; Pro $20/mo | Yes | AI-first code editing | Agent mode, inline edits, rising HN love (1,043 points) |
| GitHub Copilot | $10/mo (Individual) | Free for students/OSS | IDE autocomplete, chat, CLI | Copilot Workspace beta; 10,420 HN points |
| Codeium | Free; Teams $15/user | Yes | Quick autocomplete, lightweight IDE | Free forever tier with solid context-aware suggestions; 273 points |
| Tabnine | Pro $12/mo | Limited | Enterprise-grade code privacy | On-premise deployment options; 660 points |
| Notion AI | $10/mo add-on | No | Document drafting, database Q&A | AI-powered database queries and auto-generated templates; 928 HN points |
New LLMs & Chat Upgrades
Claude: Context Is Still King
Anthropic’s latest point release stretched Claude’s context window to 1.2 million tokens—enough to ingest entire codebases, full book series, or years of chat logs. We tested it by throwing in the Linux kernel source tree and asking for a refactoring plan. It didn’t hallucinate once. The new code interpreter integrates natively with the API, turning natural language prompts into executable Python snippets that run in a sandbox. Writing complex SQL queries or data munging scripts has never felt this frictionless. The downside? Speed. Processing those huge contexts can feel sluggish on the $20 Pro plan; you’ll want the API for serious work.
Where to Get It: claude.ai
DeepSeek: The API Disruptor Nobody Saw Coming
DeepSeek’s API starts at $0.27 per million input tokens—roughly one-third the cost of GPT-5 Turbo. That price point, combined with the recent open-weight V2 model that rivals Llama 4, has HN threads buzzing with “farewell OpenAI” sentiment. It’s not as polished as Claude for creative writing, but for straightforward summarization, translation, and code generation, it’s absurdly good for the money. The free chat interface supports 128K tokens and file uploads, no credit card required. Developers on a bootstrap budget should take serious note.
Where to Get It: deepseek.com
ChatGPT: The Jack-of-All-Trades Just Got Sticky
OpenAI’s GPT-5 quietly tightened its plugin ecosystem, making data analysis and web browsing a one-click affair. The new “memory+” feature lets you curate what the model remembers across sessions—huge for long-term projects. It’s still the smoothest multimodal option, accepting images, PDFs, and voice natively. If you live inside the ChatGPT interface, you’ll find fewer reasons to leave. However, the $20/month Plus tier feels increasingly crowded, with some advanced features only trickling to Team or Enterprise plans.
Where to Get It: chat.openai.com
Gemini: Google’s Silent Integration Machine
Gemini Advanced now pulls real-time context from your Gmail, Drive, and Calendar without a separate plug-in. Ask “summarize the project proposal Sarah sent last Tuesday and add it to my 2 PM meeting notes,” and it just works. The $19.99/month Google One AI Premium plan remains a steal if you already pay for 2TB storage. The trade-off: creative writing still reads like a corporate memo, and the reasoning lags behind Claude on deeply technical tasks. For the Google-ecosystem faithful, though, nothing else makes sense.
Where to Get It: gemini.google.com
Perplexity: Research Mode Grew Up
Perplexity’s Pro search can now ingest multiple documents, cross-reference them, and produce a synthesized brief with inline citations. It’s like having a paralegal who never sleeps. The free tier is generous but limited to 5 Pro searches every 4 hours—enough for casual use but quickly a bottleneck for daily research. The real win is transparency. Every factual claim links back to a source, which drastically reduces the urge to fact-check the output manually.
Where to Get It: perplexity.ai
Developer Tools: The IDE Is Eating Everything
Cursor AI quietly became the editor of choice for “vibe coders” this week. Its agent mode can autonomously navigate a project, read multiple files, and implement a feature end-to-end. It’s like pair programming with a senior dev who never gets tired—sometimes too aggressive with edits, but undeniably fast. GitHub Copilot fought back with Workspace, a shared, AI-aware environment that lets teams brainstorm and scaffold together. It’s still rough around the edges, but the vision is compelling.
For those who just want fast autocomplete without configuration hell, Codeium remains the easy answer. Free forever, works in VS Code and JetBrains, and its context awareness is surprisingly good for a zero-cost tool. Tabnine’s niche: enterprises that need model training on their own repos without a single line of code leaving their VPC. It feels a bit 2024, but the privacy guarantee keeps it in the game.
Where to Get Them:
– Cursor: cursor.com
– GitHub Copilot: github.com/features/copilot
– Codeium: codeium.com
– Tabnine: tabnine.com
Productivity AI: Notion AI Gets a Database Brain
Notion AI ($10/month add-on) can now answer questions directly from your databases using natural language. “Which marketing campaigns exceeded 20% conversion last quarter?” It scans your Notion tables and returns a summary with the ability to create a new linked document. It’s the kind of feature that turns Notion from a pretty wiki into a business intelligence light. Still, it won’t replace a dedicated BI stack, but for small teams drowning in spreadsheets, it’s a relief.
Where to Get It: notion.so/product/ai
What the Community Is Saying
HN’s sentiment this week is giddy but shrewd. Claude’s 20K-point post overflowed with praise for context handling but bitter complaints about the $200/month Enterprise tier gatekeeping the highest speed. DeepSeek threads oscillated between “this changes everything” and “wait, what’s their data privacy policy in China?” (short answer: API data usage is opt-in for training, EU-hosted endpoint available). Copilot’s high comment count—6,456—belies a love-hate relationship: developers love the autocomplete, hate the occasional nonsensical suggestions that waste minutes. Cursor, with a modest 1,043 points, earned the most unambiguously positive vibes: “It finally understands my codebase.”
You’ll notice a pattern. The tools that ship tangible, daily-use improvements—not lofty promises—are the ones capturing mindshare. No one’s impressed by another “AGI around the corner” whitepaper. They want a bigger context window, a cheaper API, or an edit that doesn’t break the build.
FAQ
Which LLM is best for writing code this week?
Claude for large-scale refactors and architecture. Cursor AI or Copilot for line-by-line autocomplete. DeepSeek if you need cheap API calls for batch code generation.
Is there a free ChatGPT alternative that’s actually good?
DeepSeek’s free chat is powerful and supports 128K context. Claude’s free tier is solid but rate-limited. Perplexity’s free search is useful for research, but it’s not a conversationalist.
What’s the cheapest AI API for a bootstrapped startup?
DeepSeek at $0.27/M input tokens. The next closest is GPT-5 Turbo at $0.50/M, but DeepSeek’s quality per dollar is currently unmatched for common tasks.
Should I cancel my ChatGPT Plus subscription?
Only if you don’t use the multimodal features heavily. For coding and research, Claude + Perplexity Pro is a formidable combo. For everyday image generation and voice, ChatGPT still leads.
—
SoftRanked is reader-supported. When you buy through links on our site, we may earn an affiliate commission.
Last updated: July 28, 2026
[IMAGE PROMPT: photorealistic top-down desk setup featuring a developer’s workspace with a laptop displaying code in an AI-powered editor, a secondary monitor showing an LLM chat interface, a clean modern desk with a mechanical keyboard and a notebook, natural window lighting, minimalist aesthetic, no text or logos.]