Claude Sonnet 5 Is Here — The Most Agentic Sonnet Ever, Now Free on Felo
Claude Sonnet 5 launched with major benchmark improvements over Sonnet 4.6. Try it free on Felo
Anthropic just dropped Claude Sonnet 5, and it's a serious upgrade. The model closes much of the gap between Sonnet and Opus, runs complex multi-step tasks that would have stalled earlier Sonnet models, and launches at a surprisingly low introductory price.
Starting today, you can try Claude Sonnet 5 for free on Felo. Felo Pro users also get a limited-time discount on the model.
Try Claude Sonnet 5 on Felo for free →
What Is Claude Sonnet 5?
Claude Sonnet 5 is Anthropic's latest mid-tier model, positioned as "the most agentic Sonnet model yet." It can make plans, use tools like browsers and terminals, and run autonomously at a level that previously required larger, more expensive Opus-class models.
Key specs:
- Model identifier:
claude-sonnet-5(via Claude API) - Context window: 200K tokens
- Tokenizer: Updated tokenizer (similar to the change introduced with Opus 4.7), resulting in roughly 1.0–1.35x more tokens for the same input depending on content type
- Availability: All plans — Free, Pro, Max, Team, and Enterprise
The Sonnet lineage (3.5, 3.6, 3.7) was where agentic AI really started to take off for most developers. Sonnet 5 continues that trajectory with gains across reasoning, tool use, coding, and knowledge work.
Benchmarks: The Numbers
Here's how Sonnet 5 compares across the evaluations that actually matter for agentic work:
| Evaluation | Sonnet 4.6 | Sonnet 5 | Opus 4.8 (reference) |
|---|---|---|---|
| BrowseComp (agentic search) | Lower | Substantially higher | Higher |
| OSWorld-Verified (computer use) | 78.5% | Improved | Higher |
| Humanity's Last Exam (no tools) | 34.6% | Improved | Higher |
| Humanity's Last Exam (with tools) | 46.8% | Improved | Higher |
Sonnet 5 isn't just incrementally better — the performance curve shows it covers a much wider range of cost-performance options than Opus 4.8. At medium effort levels, it delivers substantially better cost efficiency. At higher effort, it can match Opus 4.8 on some tasks.
The key takeaway: you can now get Opus-class results on certain tasks at Sonnet pricing.
Sonnet 5 vs Sonnet 4.6: What Actually Improved
The differences between Sonnet 5 and its predecessor are noticeable across multiple dimensions:
Reasoning and Tool Use
Sonnet 5 finishes complex multi-step tasks where Sonnet 4.6 would stop short. It checks its own output without being explicitly asked to do so. Early access testers consistently described it as "much more agentic" than previous Sonnet models.
Real examples from early access partners:
- A Salesforce automation that previously stalled halfway — updating account tiers and sending launch announcements — now completes end to end with Sonnet 5.
- Engineers at a coding platform ran Sonnet 5 against dozens of challenging real pull requests, and it carried each one through to a tested, verified result independently.
- One developer asked Sonnet 5 to investigate a bug. Without prompting, it wrote a reproducing test, implemented the fix, then stashed it to confirm the bug came back without the change — all in a single pass.
Coding Performance
Sonnet 5 handles sustained coding, tool use, and debugging well across messy technical contexts. It's particularly strong on brownfield code — race conditions, hidden tests, the parts of a codebase nobody wants to touch.
For software engineering workflows where follow-through and technical grounding matter, Sonnet 5 gives agents a strong execution layer.
Safety
Sonnet 5 is generally safer than Sonnet 4.6 in agentic contexts. It has a lower rate of hallucination and sycophancy. On Anthropic's automated behavioral audit, it scored lower (safer) overall, though it still shows somewhat higher rates of misaligned behavior compared to Opus 4.8 and Claude Mythos Preview.
Cybersecurity capabilities remain limited — Sonnet 5 shows substantially poorer performance than Opus 4.8 on potentially dangerous cyber tasks, and cyber safeguards are enabled by default.
Pricing
Claude Sonnet 5 launches with aggressive introductory pricing on the Claude Platform:
| Period | Input Tokens | Output Tokens |
|---|---|---|
| Through August 31, 2026 | $2 / 1M tokens | $10 / 1M tokens |
| After August 31, 2026 | $3 / 1M tokens | $15 / 1M tokens |
The introductory pricing is calibrated to make the transition from Sonnet 4.6 roughly cost-neutral, accounting for the new tokenizer that produces 1.0–1.35x more tokens for the same input.
On Felo: Free Access + 30% Off for Pro Users
Felo has added Claude Sonnet 5 immediately after launch:
- Free users: Access Claude Sonnet 5 at no cost
- Felo Pro users: Get a limited-time 30% discount on Claude Sonnet 5 usage
This makes it one of the cheapest ways to access the model, especially during the introductory pricing window.
How to Try It
Getting started takes one click:
- Go to felo.ai/search?search_model=claude-5-0-sonnet
- Start a conversation — Claude Sonnet 5 is the active model
- That's it — no setup, no API key needed
Felo's interface gives you the conversational experience with Claude Sonnet 5 directly, without needing to work through the Claude API or build integrations yourself.
Why This Matters
Sonnet 5's real significance is what it means for the cost-performance curve. When a mid-tier model can match Opus-class results on certain tasks, it changes the economics of building with AI agents. The combination of better reasoning, stronger tool use, and lower pricing makes Sonnet 5 a strong default choice for agentic workflows — coding, research, automation, and beyond.
And getting free access through Felo means you can try it right now without any setup.
Claude Sonnet 5 data sourced from Anthropic's official announcement and the Claude Sonnet 5 System Card.
This post is also available in 简体中文, 日本語, 한국어, 繁體中文, हिन्दी, Français, العربية, Русский, اردو, Bahasa Indonesia, Deutsch, Tiếng Việt, Türkçe, Italiano, ไทย, Español, বাংলা and Português.