Claude Sonnet 5 has been the default model for every free and Pro Claude user since 30 June 2026, which means most people reading this have already used it without knowing its name. This review covers what actually changed from Sonnet 4.6, what it costs now that the introductory pricing has expired, and — since Anthropic has shipped three more Claude tiers since this one launched — where Sonnet 5 fits in the current lineup.
By Shekhar Chandran | Published 5 September 2026 | 7 min read | Pricing and benchmark figures verified against Anthropic’s official Claude pricing page and independent Artificial Analysis testing.
📅 5 September 2026 · 🕐 7 min read · 🗂️ AI Updates
This page covers:
- What Claude Sonnet 5 actually improved over Sonnet 4.6
- Current pricing, now that the introductory rate has expired
- How it compares to Claude Opus 5 and to GPT-6 Astra
- Whether it’s still worth using now that Fable 5.1 and Mythos 5.1 exist
This page does not cover: Anthropic’s flagship model in depth — see our Claude Opus 5 review for that, or the full Claude AI in India guide for the whole model lineup and India pricing.
Jump to a Section
- What is Claude Sonnet 5?
- How much does Claude Sonnet 5 cost now?
- How does Sonnet 5 perform?
- Where does Sonnet 5 fit in Anthropic’s current lineup?
- Claude Sonnet 5 in India
- AIInsider Verdict
- Quick Answers
What is Claude Sonnet 5?
Claude Sonnet 5 (model ID claude-sonnet-5) is Anthropic’s mid-tier model, released on 30 June 2026 as the replacement for Sonnet 4.6. It’s the model most people mean when they say “Claude” without specifying which one — it’s the default for Free and Pro plans on claude.ai, and it’s also available on Max, Team, Enterprise, Claude Code, and the API.
The headline change from Sonnet 4.6 is agentic reliability, not raw intelligence. Anthropic built it to plan a multi-step task, execute it with less hand-holding, and catch its own mistakes without being explicitly asked to check its work. It ships with a native 1M-token context window built directly into Claude Code, rather than the API-only beta tier Sonnet 4.6 required.
One detail that trips people up when they migrate: Sonnet 5 uses an updated tokenizer that counts roughly 1.0–1.35x more tokens for the same text as Sonnet 4.6 did. Same rate card, more tokens per task — worth checking your own workload before assuming costs stay flat.
How much does Claude Sonnet 5 cost now?
Sonnet 5 launched with introductory API pricing of $2 per million input tokens and $10 per million output tokens. That window closed on 31 August 2026 — five days before this article was published — so every Sonnet 5 API call today runs at the standard rate, confirmed on Anthropic’s official pricing documentation.
| Tier | Input (per 1M tokens) | Output (per 1M tokens) | Status |
|---|---|---|---|
| Introductory | $2.00 | $10.00 | Expired 31 Aug 2026 |
| Standard (current) | $3.00 | $15.00 | Active now |
That standard rate is identical to what Sonnet 4.6 charged — Anthropic didn’t raise the sticker price for a meaningfully stronger model. Combined with the tokenizer change above, real per-task spend can land higher than Sonnet 4.6 even at the same rate, so budget from actual token counts rather than the old model’s invoices.
🇮🇳 Indian developers building on the API — at a mid-market rate of ₹94.42/$1, standard pricing works out to roughly ₹283 per million input tokens and ₹1,416 per million output tokens. For chat use rather than API use, Claude’s free and Pro plans (covered with full India pricing in our Claude AI Pricing India guide) already include Sonnet 5 at no extra cost over what you’re likely already paying.
How does Sonnet 5 perform?
On SWE-bench Pro, Anthropic’s own disclosed coding benchmark, Sonnet 5 scores 63.2% — ahead of Sonnet 4.6’s 58.1%, though still behind Opus 4.8’s 69.2% at the time both were current, per Anthropic’s official launch announcement. Anthropic never published a directly comparable SWE-bench Verified number for Sonnet 5, so treat any site claiming to know it as guessing.
The more interesting numbers are on agentic tasks rather than raw coding. On BrowseComp (agentic web search) and OSWorld-Verified (computer use), Sonnet 5 approaches Opus 4.8’s scores at medium reasoning effort — a genuinely unusual result for a mid-tier model closing that much of the gap to the flagship above it. Independent testing from Artificial Analysis largely corroborates this, placing Sonnet 5 ahead of Opus 4.8 specifically on agentic knowledge-work benchmarks, even as it trails on their broader Intelligence Index.
Anthropic also reports a lower rate of undesirable behaviours than Sonnet 4.6, with cyber safeguards enabled by default rather than opt-in.
My take: the benchmark that actually matters here isn’t the coding score — it’s that Sonnet 5 closes so much of the gap to Opus-class computer use at a third of the price. If your workload is browser automation or tool-calling rather than deep single-shot coding, Sonnet 5 is doing something genuinely different from a normal mid-tier refresh.
Where does Sonnet 5 fit in Anthropic’s current lineup?
This is the part most reviews miss, because they were written the week Sonnet 5 launched — before Anthropic shipped anything above it. As of this article, the live Claude lineup looks like this:
| Model | Positioning | Standard price (per 1M tokens) |
|---|---|---|
| Claude Haiku 4.5 | Fast, cheap | $1 / $5 |
| Claude Sonnet 5 | Everyday default, agentic workhorse | $3 / $15 |
| Claude Opus 5 | Maximum capability, complex reasoning | Higher — see Opus 5 review |
| Claude Fable 5.1 / Mythos 5.1 | Frontier tier, restricted access | Not broadly available |
Sonnet 5 hasn’t been superseded within its own tier — there’s no Sonnet 5.1 as of this writing, despite persistent rumours since February. It’s still the current, actively-served default model for the vast majority of Claude users. What’s changed is the context around it: Opus 5 now sits above it as the flagship (Opus 4.8’s benchmark numbers cited above are the comparison point Sonnet 5 launched against, not necessarily where Opus 5 sits today), and the Fable/Mythos tier exists above both, with its own access restrictions we’ve covered separately.
Claude Sonnet 5 in India
Nothing about Sonnet 5’s rollout was India-specific — it shipped simultaneously worldwide as the default model, the same way every Claude release has. The practical India-specific questions (UPI payment gaps, INR billing, which plan to buy) are covered in full in our dedicated Claude Pricing India guide and our guide to paying for Claude Pro from India — both still current, since none of the payment mechanics changed with this release.
AIInsider Verdict
Score: 8.2/10. Sonnet 5 is a genuinely strong mid-tier model that closes an unusual amount of ground toward flagship-level agentic performance without a flagship price tag. It’s not the newest thing Anthropic has shipped, but it’s still the model doing the actual work for most Claude users today.
What we liked:
- Approaches Opus 4.8-level computer use and agentic search at a third of the price
- Native 1M-token context window built into Claude Code by default
- Self-verification that catches mistakes without explicit prompting
- Standard pricing unchanged from Sonnet 4.6 despite the capability jump
What holds it back:
- Introductory pricing already expired — you’re paying standard rates now
- Updated tokenizer means real spend can exceed Sonnet 4.6 even at the same rate card
- Already two tiers old relative to Anthropic’s current frontier (Fable 5.1 / Mythos 5.1)
- No independently-verified SWE-bench Verified score, only Anthropic’s own SWE-bench Pro number
Our recommendation, plainly: if you’re already on Claude Free or Pro, you’re using Sonnet 5 by default and don’t need to do anything. If you’re building on the API and weighing model tiers, Sonnet 5 is the right default for agentic and tool-calling work; reserve Opus 5 for tasks that genuinely need maximum reasoning depth rather than agentic reliability. Don’t wait for a “Sonnet 5.1” — nothing suggests one is imminent.
Quick Answers
When was Claude Sonnet 5 released?
30 June 2026, replacing Sonnet 4.6 as the default model for Claude Free and Pro plans.
How much does Claude Sonnet 5 cost now?
Standard pricing is $3 per million input tokens and $15 per million output tokens, the same as Sonnet 4.6. The $2/$10 introductory rate expired on 31 August 2026.
Is Claude Sonnet 5 better than Claude Opus 5?
Not for raw reasoning depth — Opus is Anthropic’s flagship for that. Sonnet 5 is optimized for agentic execution, tool use, and computer-use tasks at a fraction of Opus pricing.
Does Claude Sonnet 5 have a 1 million token context window?
Yes, natively, built into Claude Code by default — unlike Sonnet 4.6, which offered 1M tokens only as an API-only beta tier.
Is there a Claude Sonnet 5.1?
No, not as of this article. Sonnet 5 remains Anthropic’s current Sonnet-tier model despite recurring rumours of a point release.
Related Articles
- Claude AI in India: The Complete 2026 Guide — the full model lineup and India-specific pricing.
- Claude Opus 5 Review India (2026) — Anthropic’s flagship model, reviewed.
- Claude AI Pricing India 2026 — official INR rates across every plan.
- ChatGPT Astra (GPT-6) Review — how OpenAI’s current flagship compares.
Published 5 September 2026. Pricing and benchmark figures verified against Anthropic’s official Claude pricing page; USD–INR rate checked 3 September 2026. AIInsider.in is independent and not affiliated with Anthropic.