Anthropic Releases Opus 4.8 With Fast Mode Three Times Cheaper, Standard Price Unchanged

Welcome back to SavvyMonk, your one-stop for AI and tech news that actually matters.

Anthropic shipped a new flagship model yesterday, and the turnaround was unusually fast.

Let's get into it.

You prompt. The agent builds. Then it asks for a database.

Ghost is postgres made for this. Spin one up in seconds. Fork it like a branch. Delete it when you're done. Pay nothing when it's idle.

Your agent gets full sql, mcp support, as many databases. No dashboards. No provisioning. No forgotten dev databases draining your card at month end.

Build a weekend app. Fork the schema three different ways. Throw two of them out. Ghost doesn't care. The next prompt can spin up a fresh one.

You're already vibe-coding the app. Stop wiring up the backend.

Unlimited databases. Unlimited forms. 100 compute/hrs a month. 1tb of storage. Free.

Download Ghost

Anthropic released Claude Opus 4.8 on May 28, only 41 days after Opus 4.7 landed. That is roughly six weeks between flagship releases. The most recent Sonnet and Haiku models, by comparison, are three and seven months old.

Opus 4.7 had a rough reception among developers, with users publicly calling it a step back in reliability. Anthropic moved quickly, and it is hard not to read the 41-day cycle as a direct response to that feedback.

It did not help that OpenAI and Google both shipped notable new model releases in that same window, raising the competitive pressure.

What Changed

The most significant improvement in Opus 4.8 is its honesty. Anthropic trained the model to flag uncertainty in its own work rather than glossing over problems.

It’s already smarter than its previous models. It passed the “Car Wash Test.”

By their own evaluation, Opus 4.8 is around four times less likely than its predecessor to let flaws in code pass without remarking on them. Early testers found it more willing to push back on a plan it thinks is unsound, and more likely to catch its own mistakes before a developer has to find them.

On benchmarks, agentic coding improved from 64.3% to 69.2% between 4.7 and 4.8. Multidisciplinary reasoning with tools climbed from 54.7% to 57.9%. For browser agent work, the model scored 84% on Online-Mind2Web, surpassing both Opus 4.7 and OpenAI's GPT-5.5. It also became the first model to complete every case end-to-end on the Super-Agent benchmark at parity cost to GPT-5.5, and the first to break 10% on the all-pass standard of the Legal Agent Benchmark.

On alignment, Anthropic's team found the model reaches new highs on prosocial traits like supporting user autonomy and acting in the user's best interest. Rates of misaligned behavior are substantially lower than in Opus 4.7, and comparable to Claude Mythos Preview, Anthropic's highest-capability model.

Three New Features Shipping Alongside

The first is dynamic workflows, available in research preview inside Claude Code for Enterprise, Team, and Max plan users. It lets Claude plan a large task and then spin up hundreds of parallel subagents in a single session to execute it. Anthropic's example use case is a codebase-scale migration across hundreds of thousands of lines of code, from planning through to merge, using the existing test suite as the quality bar.

The second is effort control on claude.ai and Claude Cowork, available across all plans. A new control next to the model selector lets users decide how hard Claude should work on a given response. Higher effort triggers more frequent and deeper thinking. Lower effort trades response quality for speed and conserves rate limit capacity.

The third is a change to the Messages API. Developers can now insert system-level instructions inside the messages array mid-task, without breaking the prompt cache or routing the update through a user turn. For long-running agentic workflows, that reduces friction in how developers manage changing context over time.

Pricing and Availability

Pricing is unchanged from Opus 4.7. Regular usage stays at $5 per million input tokens and $25 per million output tokens. Fast mode, where the model runs at 2.5 times its normal speed, is priced at $10 per million input tokens and $50 per million output tokens.

Fast mode itself is now three times cheaper than it was for prior models, which matters for developers building latency-sensitive products. The model is accessible via the API string claude-opus-4-8 and is available across Claude Pro, Max, Team, and Enterprise plans, as well as on AWS, Google Cloud, and Microsoft Foundry.

What's Next

Anthropic also noted in the release post that Claude Mythos Preview, currently limited to a small number of organizations doing cybersecurity work, is expected to open to all customers within the coming weeks once additional safety safeguards are in place.

The Bottom Line

Opus 4.8 is a direct course correction after Opus 4.7 underdelivered. The 41-day release cycle shows Anthropic can move faster when the situation demands it. The honesty improvements are the most practically significant change for anyone building on the model in production, since an AI that flags its own errors is far more reliable in automated workflows where no human is watching every step. At the same price, it is a straightforward upgrade.

Category: Developer Productivity

"You are reviewing code I have written. Go through it carefully and flag every place where you are uncertain the logic is correct. Do not explain what the code does. Focus only on potential bugs, edge cases, or assumptions that could fail in production. Be direct if something looks wrong even if you are not certain. Here is the code: [paste your code]"

The most telling improvement in Opus 4.8 is not the benchmark scores. It is that the model will now tell you when it is not sure. In automated agent workflows where no human is checking every step, that matters more than raw performance numbers. A model that overclaims is a liability you have to manage. A model that flags uncertainty is something you can actually build on. Hit reply, I read every response.

See you in the next one.

— Vivek

P.S. If you know developers or AI engineers who want a clean read on what is actually shipping in AI, pass this along. They can subscribe at https://savvymonk.beehiiv.com/