Claude Upload Limit, Code Router, Claude SEO Tool, and Sonnet vs Opus: Everything You Need to Know in 2026
Four questions come up repeatedly from Claude users in 2026 — about uploading files, routing between models, using Claude for SEO, and choosing between Sonnet and Opus. This article answers all four with verified, real data. No padding, no guesswork — just what you actually need to know to use Claude more effectively right now.
Claude Upload Limit: Exact Numbers You Should Know
The Claude upload limit is 30 MB per file with a maximum of 20 files per conversation. This applies consistently across free and paid plans in the claude.ai chat interface. The file size constraint doesn’t change between plans — what changes is context window capacity and message frequency.
Here’s the complete breakdown by environment:
| Environment | Per File Limit | File Count |
|---|---|---|
| Claude Chat (all plans) | 30 MB | 20 per conversation |
| Claude Projects (Pro/Team) | 30 MB | Unlimited |
| API Messages | 32 MB | — |
| Files API (Beta) | 500 MB | — |
| Amazon Bedrock | 4.5 MB (docs) | 5 per request |
Supported File Formats
Claude accepts 10 document formats — PDF, DOCX, TXT, CSV, HTML, RTF, EPUB, JSON, ODT — and 4 image types: JPEG, PNG, GIF, and WebP. Images can be up to 8,000 × 8,000 pixels. For Excel (XLSX), you need to enable the “Analysis Tool” in your account settings first — many users miss this step and get confused when uploads fail.
PDFs under 100 pages get full multimodal analysis — Claude reads both text and visual elements like charts and diagrams. PDFs over 100 pages drop to text-only extraction. This matters if you’re uploading dense reports expecting Claude to read embedded graphs.
The Real Bottleneck: Context Tokens, Not File Size
The 30 MB file size limit is not usually the problem. The real constraint is Claude’s 200,000-token context window. A text-heavy 5 MB PDF can consume more tokens than a 25 MB image-heavy document. When five large PDFs fill up the context window simultaneously, Claude silently starts dropping earlier content to make room — and you won’t notice until it misses details you uploaded earlier.
Practical fix: For large document sets, use Claude Projects (Pro/Team plan). Files persist across conversations and Claude pulls only what’s needed per query, which stretches your effective context capacity significantly. For developers needing to reference the same large file repeatedly, the Files API (beta since April 2025) stores files server-side at up to 500 MB and lets you reference them by ID — no re-uploading required.
Claude Code Router: What It Is and Why Developers Use It
The Claude Code Router is an open-source proxy tool that sits between Claude Code’s terminal interface and AI model backends. Instead of routing every request directly to Anthropic’s API, it intercepts requests and redirects them to whichever model you configure — OpenRouter, DeepSeek, Ollama (local models), Google Gemini, or 8+ other providers.
The project, maintained by GitHub user @musistudio, has accumulated over 26,000 GitHub stars and 2,100+ forks as of early 2026. It’s not a hobby experiment — it’s widely used production tooling.
In 2026, “Claude Code Router” actually refers to two distinct things you should understand separately:
- Built-in routing (opusplan): Anthropic’s own model alias inside Claude Code that automatically uses Claude Opus 4.6 for planning and architectural reasoning, then switches to Claude Sonnet 4.6 for code execution. Zero setup — one command enables it. Best for developers who want cost optimization within the Anthropic ecosystem.
- Community router (CCR): The open-source proxy by @musistudio. Routes Claude Code requests to non-Anthropic providers — useful if you want to run DeepSeek for routine tasks, use local Ollama models for sensitive code, or simply reduce your monthly API spend.
Why Developers Actually Use It
The core problem it solves: not every coding task needs Claude Sonnet or Opus. Renaming variables, writing basic tests, generating boilerplate — cheaper or free models handle these fine. The Router lets you route those tasks to DeepSeek ($0.14 per million tokens) or a local Ollama model, while keeping expensive Claude models for the complex work.
Real cost impact is $40–$150 per month in savings for a typical developer workflow, according to independent testing published in February 2026. The “10x savings” marketing is overstated — expect 3x to 5x realistically with a mixed routing setup costing $15–40 per month. The tool itself costs nothing (MIT license).
The catch: some providers don’t support tool-calling, which means file edits, git commands, and terminal operations can break. And you lose Claude’s Constitutional AI safety guardrails when routing to non-Anthropic models. For most developers doing standard development work, neither issue is a dealbreaker. For security-sensitive or compliance-constrained environments, stick with Anthropic’s native models.
Quick Setup (3 Steps)
- Install the router:
npm install -g @musistudio/claude-code-router - Create
~/.claude-code-router/config.jsonand add your providers and routing rules - Run
ccr startin one terminal,ccr codein another — Claude Code now routes through your config
GitHub repository: github.com/musistudio/claude-code-router
Claude as an Claude SEO Tool: What It Actually Does Well
Claude is not a dedicated SEO platform — it doesn’t pull live keyword data, check domain authority, or run technical crawls. What it is exceptionally good at is the thinking and writing work that SEO requires: content strategy, semantic optimization, title testing, on-page structuring, and content that satisfies Google’s Helpful Content guidelines.
Here’s where Claude genuinely earns its place in an SEO workflow:
Content Writing and On-Page Optimization
Claude writes in natural, semantically rich English that Google’s NLP systems score well on E-E-A-T signals. It structures content with logical heading hierarchies, writes in active voice, avoids keyword stuffing patterns, and maintains topical depth across long-form pieces. For content-heavy sites that publish frequently, Claude’s ability to hold brief constraints across a full 3,000-word article — consistently — saves substantial editing time.
Where to use it: blog posts, service pages, product descriptions, FAQ sections, and any content where depth, readability, and semantic alignment matter more than technical crawl data.
Title and Meta Description Testing
Feed Claude your target keyword, the page’s intent, and character limits. Ask for 10 title variations with different emotional hooks — curiosity, urgency, specificity, benefit-led. Then test the top 3 in Google Search Console. This workflow consistently produces better CTR than using a single internally-generated title, and Claude’s understanding of intent means it rarely produces titles that don’t match page content.
Keyword Clustering and Content Gap Analysis
Given a seed keyword and a content brief, Claude organizes related keywords into semantic clusters and suggests heading structures that cover the topic comprehensively. This is the manual work that SEO tools do mechanically — Claude does it with actual understanding of why certain subtopics belong together. For site owners without a full SEMrush or Ahrefs subscription, this is a practical alternative for content planning.
Competitor Analysis and Structure Review
Paste a competing article’s full text or structure into Claude and ask it to identify what the article covers that yours doesn’t, what the likely search intent is, and where your draft is thinner. Claude’s 200,000-token context window lets you work with full competitor content, full drafts, and full briefs simultaneously — a workflow that’s genuinely difficult to replicate with most alternatives.
Use alongside: Google Search Console (for real performance data), Ahrefs or SEMrush (for keyword volume and competitor backlink data), and RankMath or Yoast (for technical WordPress on-page checks). Claude fills the strategy and writing layer — not the data layer.
Claude Sonnet 4.6 vs Opus 4.6: The 2026 Comparison That Actually Matters
This is the Claude question most developers and professionals search for in early 2026 — and the answer has changed significantly compared to previous model generations. The gap between Sonnet and Opus has narrowed to the point where the choice is now primarily economic, not qualitative.
The Benchmark Numbers
| Benchmark | Sonnet 4.6 | Opus 4.6 |
|---|---|---|
| SWE-bench Verified (coding) | 79.6% | 80.8% |
| OSWorld-Verified (computer use) | 72.5% | 72.7% |
| GDPval-AA (office tasks) | 1633 Elo | 1606 Elo |
| GPQA Diamond (research reasoning) | 74.1% | 91.3% |
| Terminal-Bench 2.0 | ~59% | 65.4% |
| Max output tokens | 64K | 128K |
| Speed (tokens/sec) | 40–60 t/s | 20–30 t/s |
| API pricing (input/output per 1M tokens) | $3 / $15 | $15 / $75 |
The number that reframes this decision entirely: on GDPval-AA — the benchmark measuring real-world office and knowledge work — Sonnet 4.6 outscores Opus 4.6 at 1633 vs 1606 Elo. For content creation, structured writing, financial task automation, and office work, Sonnet isn’t just comparable — it’s ahead.
Where Opus 4.6 Still Wins
Opus holds a clear advantage in three specific areas: deep scientific and expert reasoning (91.3% on GPQA Diamond vs Sonnet’s 74.1%), very large codebases requiring full-repository context and coherent multi-file changes, and multi-agent coordination through Claude Code’s Agent Teams feature. Its 128K output ceiling versus Sonnet’s 64K also matters for tasks requiring massive single-response artifacts — full legal documents, complete system designs, comprehensive research reports.
The Decision Framework
- Use Sonnet 4.6 for: daily coding, bug fixes, content writing, SEO work, financial analysis, computer use, high-volume API calls, automated pipelines, and anything where speed or cost efficiency matters. Handles 80–90% of professional workloads at 5x lower cost.
- Use Opus 4.6 for: research and expert domain reasoning, large codebase refactoring (10K+ lines), multi-agent coordination, deep security audits, tasks requiring 128K output, and high-stakes analysis where failure is expensive.
The most efficient approach in 2026 is the hybrid: Sonnet 4.6 as the default for virtually everything, Opus 4.6 as the escalation layer for the 10–20% of tasks where its ceiling genuinely matters. Companies doing this report 60–80% reduction in API costs with no meaningful drop in output quality for the majority of their workflows.
Quick Reference: All Four Topics at a Glance
- Claude upload limit: 30 MB per file, 20 files per chat. Projects (Pro) = unlimited files. Files API (beta) = up to 500 MB. Context window (200K tokens) is the real cap to watch.
- Claude Code Router: Open-source proxy (26K+ GitHub stars) that routes Claude Code requests to cheaper or local models. Built-in opusplan alias uses Opus for planning + Sonnet for execution. Real savings: $40–$150/month. Free tool, MIT license.
- Claude for SEO: Strong for content writing, semantic structuring, title testing, and keyword clustering. Pair with GSC + Ahrefs/SEMrush for data. Claude handles strategy and content quality — not crawl analytics.
- Sonnet vs Opus: Sonnet 4.6 beats Opus on office tasks (1633 vs 1606 Elo), matches it on computer use (72.5% vs 72.7%), and costs 5x less. Use Opus only for deep research, massive codebases, or multi-agent coordination.
Frequently Asked Questions
What is the Claude file upload limit in 2026?
Claude’s file upload limit is 30 MB per file with a maximum of 20 files per conversation in the chat interface. In Claude Projects (Pro/Team plans), files can be unlimited in number at the same 30 MB per-file cap. The Files API (beta) supports individual files up to 500 MB. On Amazon Bedrock, the limit drops to 4.5 MB per document with a maximum of 5 documents per request.
What is Claude Code Router and is it free?
Claude Code Router is an open-source proxy tool (MIT license) that intercepts Claude Code CLI requests and routes them to alternative AI providers including DeepSeek, Google Gemini, OpenRouter, Ollama, and others. It’s completely free to use — you only pay for whichever AI model API you route requests to. It has 26,000+ GitHub stars as of early 2026 and is actively maintained by @musistudio. Real monthly savings for typical developer workflows range from $40–$150.
Can I use Claude as an SEO tool?
Yes, with the right expectations. Claude excels at SEO content writing, semantic keyword organization, heading structure, meta title generation, competitor content analysis, and producing E-E-A-T-optimized copy. It does not provide live keyword volume data, domain authority metrics, or technical crawl analysis. Use Claude for the strategy and content layer, and pair it with tools like Google Search Console, Ahrefs, or SEMrush for data.
Is Claude Sonnet 4.6 better than Opus 4.6?
For most everyday tasks — yes, and at a much lower price. Sonnet 4.6 matches Opus 4.6 on computer use (72.5% vs 72.7%), comes within 1.2 percentage points on coding benchmarks, and actually outperforms Opus on office and knowledge work tasks (1633 vs 1606 Elo). It runs at 40–60 tokens/sec versus Opus’s 20–30, and costs $3/$15 per million tokens versus Opus’s $15/$75. For deep research, large codebase refactors, or expert scientific reasoning, Opus still leads clearly.
How do I work around Claude’s upload limit for large files?
Four practical workarounds: (1) For recurring large files, use the Files API (beta) to store them server-side at up to 500 MB each and reference by ID without re-uploading. (2) For large document sets, use Claude Projects on a Pro or Team plan — files persist across conversations and context is retrieved per query. (3) Split large PDFs into logical chunks under 30 MB before uploading. (4) Convert Excel files to CSV before upload — reduces file size and improves parsing accuracy.
Aman Alria is the founder of ClawdBot2.in and an artificial intelligence writer covering the latest AI news, tools, and trends. He breaks down complex AI topics into clear, honest content — from model comparisons and agent updates to AI regulation and learning resources. If it’s happening in AI, Aman is writing about it.