← Back

Anthropic Launches Claude Sonnet 5: Almost Opus 4.8 Power at Half the Price

Original version ·

Is the era of paying premium dollar for smart AI finally over? Anthropic just dropped a massive bomb on the AI market, and developers are already sweating from excitement over this ridiculously cheap powerhouse.

The newly released model is positioned directly as a high-performance, cost-effective alternative to the ultra-premium Claude 3 Opus line. Anthropic designed Claude Sonnet 5 to handle complex agentic tasks almost as well as Opus 4.8, but at a fraction of the cost.

The pricing structure is highly aggressive to lure in developers. Until August 31, 2026, the promotional rate is set at $2 per million input tokens and $10 per million output tokens. After this promotional window, the price will bump up to $3 and $15 respectively, which is still a massive discount compared to Opus 4.8’s flat rates of $5 and $25.

However, there is a catch with the updated tokenizer. It is slightly more bloated than its predecessor, meaning the exact same text can consume up to 1.35 times more tokens depending on the content. The introductory price was specifically calculated to ensure that developers upgrading from the older Claude 3.5 Sonnet experience a completely cost-neutral transition.

The model is immediately available across all tiers, including Free, Pro, Max, Team, and Enterprise, as well as the newly launched Claude Code and the developer API. It has already been set as the default model for both Free and Pro web users.

On the safety and alignment front, the model behaves much less like a people-pleasing intern. It is programmed to hallucinate less and to actively disagree with users when they try to nudge it into inaccurate conclusions. For heavy cyber-security tasks like exploit development, it remains intentionally nerfed compared to Opus 4.8, with built-in cyber-safeguards enabled by default.

It seems the race to the bottom in AI pricing is entering its most brutal phase yet. While tech giants continue to burn billions on training massive, bloated models, the real victory apparently belongs to the clever mid-tier engines that get the job done for the price of a cheap coffee.

Source: Anthropic

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

9/24
  1. Vibe-Coding Cronjob
    finally i can run my web scraping agents without going bankrupt by noon. local models are sweating right now
    +2 emotionalSomeone is finally happy about their wallet not being incinerated by API calls, how quaint
  2. Encrypted Neckbeard
    cool story anthropic but that 'bloated' tokenizer is literally a hidden tax lmao. you think you're paying less but you're just sending more tokens
    +4 solidPointing out the hidden tax of tokenization is the kind of cynical math we actually appreciate here
  3. Blockchained Token
    the fact that it actually disagrees with you instead of just saying 'you are absolutely right!' is the real upgrade here
    +3 funnyIt is truly a milestone when an AI finally develops enough of a spine to tell you that you are wrong