On 30/06/2026, Anthropic launched Claude Sonnet 5 — a mid-tier model the company describes as its "most agentic yet": stronger at reasoning, tool use, coding, and autonomously handling long-running tasks. Sonnet 5 becomes the default model for Free and Pro users, ships with a 1 million token context by default, and launched at an introductory price of $2 input / $10 output per million tokens, originally due to run until 31/08/2026 before rising to $3/$15. On 10/08/2026 Anthropic made the $2/$10 price permanent. For businesses, this is a move that noticeably drives down the cost of running AI agents.
Quick summary
- When: launched 30/06/2026; becomes the default model for Free + Pro (effective 01/07).
- Context: 1 million tokens by default, with up to 128K output tokens (Anthropic model docs).
- Pricing: launched as a promo of $2/$10 per Mtok until 31/08/2026; since 10/08/2026 $2/$10 is the permanent price — the planned $3/$15 never took effect.
- Quality: improved reasoning/tool use/coding over Sonnet 4.6; lower hallucination and sycophancy; better prompt-injection resistance, cyber safeguards on by default.
- Claude Code: available in Claude Code with the full 1M context on paid plans.
What happened?
According to Anthropic's official announcement, Sonnet 5 is available on every plan (Free, Pro, Max, Team, Enterprise) and via the API. The headline isn't a single benchmark but the positioning: performance approaching the flagship Opus 4.8 on many tasks at a mid-tier price — TechCrunch called it outright "a cheaper way to run agents".
Notable on the technical side: according to Anthropic's model documentation, the 1 million token context is the default (both default and max), meaning loading an entire codebase or a large document set into a single session is no longer a beta feature.
What's new on safety — notable after the Fable 5 saga
Launching just days after the US lifted the ban on Fable 5, Sonnet 5 shows Anthropic making safety a commercial priority: the company reports the model has lower hallucination and sycophancy (the tendency to "flatter" users), better resistance to prompt injection, and cyber safeguards enabled by default. For businesses putting AI into real workflows, these "hard-to-show-off" improvements are often worth more than benchmark points.
| Aspect | On Sonnet 5 |
|---|---|
| Reasoning, tool use, coding | Better |
| Hallucination & sycophancy | Lower |
| Prompt-injection resistance | Stronger |
| Cyber safeguards | On by default |
| Context | 1 million tokens by default (both default and max) |
The cost equation for businesses
VentureBeat notes that pricing Sonnet 5 well below the top model is part of the race to cut AI costs ahead of Anthropic's IPO. For API users: $2/$10 per million tokens is highly competitive for a near-flagship model — especially since agents burn tokens many times faster than regular chat.
But the familiar caveat applies: token-based pricing means costs scale with usage, and data still flows through the provider's cloud. At launch the price was billed as a promotion ending 31/08, after which it would rise to $3/$15; on 10/08/2026 Anthropic announced that $2/$10 is now permanent, and its pricing page still lists $2 input / $10 output per million tokens. A price that can change once can change again — which is exactly the point.
| Period | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| At launch (30/06/2026, billed as a promo to 31/08) | $2 | $10 |
| Planned after 31/08/2026 (cancelled on 10/08/2026) | $3 | $15 |
| Current price (permanent since 10/08/2026) | $2 | $10 |
The Namtech perspective for Vietnamese businesses
Sonnet 5 is a good choice for workloads that suit the cloud. Yet the lesson from Anthropic itself over the past three weeks (Fable 5 banned, then reinstated) still stands: cloud models change on the provider's and regulators' schedule — prices shift, models retire, access comes with conditions. For core data and processes, in-house AI on your own infrastructure remains the stable foundation: fixed costs, no dependence on promotions, and data that never leaves the organization.
Frequently asked questions
How is Claude Sonnet 5 different from Sonnet 4.6?
According to Anthropic: better reasoning, tool use, and coding; lower hallucination and sycophancy; stronger prompt-injection resistance; cyber safeguards on by default; and a 1 million token context as the default.
How much does Sonnet 5 cost?
$2 input / $10 output per million tokens. It launched as a promotion until 31/08/2026 with $3/$15 planned afterwards, but on 10/08/2026 Anthropic made $2/$10 the permanent price.
Can I use it in Claude Code?
Yes — it's available in Claude Code, with the full 1 million token context on paid plans.
What does a 1 million token context mean?
Roughly the equivalent of loading tens of thousands of pages of documents, or an entire large codebase, into one session. On Sonnet 5 this is the default context window, per Anthropic's model documentation.
Stable AI costs, no waiting for promotions
Namtech deploys in-house AI running on your infrastructure — fixed costs, data on-premises, independent of vendor pricing cycles.
Book a free consultationNote: This article was first compiled from public sources on 02/07/2026. Last updated 24/09/2026: pricing updated (the $2/$10 introductory price became permanent on 10/08/2026) and the 1M-token context sourced to Anthropic's model docs. Information is for reference and may change.