Published May 28, 2026 · Anthropic flagship model
Claude Opus 4.8 is Anthropic's new flagship AI model, and it landed on May 28, 2026 — just 41 days after Opus 4.7. For a finance and technology audience, the upgrade matters for two reasons. First, the model gets meaningfully better at coding, agentic tasks and financial analysis. Second, the price stays exactly the same. In short, you get more capability without a bigger bill.
This full guide breaks down what shipped, the benchmark numbers that count, the new pricing, and what the release means if you build products or analyze markets. New to the lineup? Start with our complete 2026 guide to Anthropic's Claude models for the full picture.
Claude Opus 4.8 is the latest version of Anthropic's most advanced publicly available model. Technically, it is a point upgrade over Opus 4.7 rather than a brand-new generation. However, the gains are concrete. Anthropic describes the model as having sharper judgment, more honesty about its own progress, and the ability to work independently for longer than before.
The headline theme is honesty. Notably, Opus 4.8 is roughly four times less likely than Opus 4.7 to let a flaw in its own code pass without flagging it. In practice, that means the model is more willing to say "I am not sure" instead of declaring a task finished too early. For anyone shipping real software or relying on AI analysis, this is a useful trait.
Anthropic published a set of self-reported benchmark scores alongside the launch. The most important signal is agentic coding, because that is where the frontier is moving fastest. Here is how the new model compares to its predecessor.
| Benchmark | Opus 4.8 | Opus 4.7 | What it measures |
|---|---|---|---|
| SWE-bench Verified | 88.6% | 87.6% | Real software bug fixing |
| SWE-bench Pro | 69.2% | 64.3% | Harder, less-contaminated coding |
| Terminal-Bench 2.1 | 74.6% | — | Command-line agent work |
| GPQA Diamond | 93.6% | — | Graduate-level reasoning |
| OSWorld (computer use) | 83.4% | 78.7% | Operating a real computer |
| GDPval-AA (Elo) | 1890 | — | Knowledge-work quality |
The biggest jump is on SWE-bench Pro, which climbs from 64.3% to 69.2%. That gain is the real coding signal, because SWE-bench Verified is close to saturation. Meanwhile, the GDPval-AA score of 1890 Elo puts Opus 4.8 about 121 points ahead of OpenAI's GPT-5.5. As always, these are Anthropic's own numbers, so treat them as a starting point and test on your own workload.
This is the part that should make finance-minded readers smile. Anthropic held standard pricing flat. Therefore, Claude Opus 4.8 costs the same as Opus 4.7.
Crucially, the new fast mode is about three times cheaper than the equivalent fast mode on previous models. As a result, latency-sensitive workloads — chat widgets, live tools, customer support — become far more affordable to run at scale. The model is available immediately across the Claude API, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. The API model ID is claude-opus-4-8.
Opus 4.8 did not arrive alone. Anthropic shipped two companion features on the same day.
Dynamic Workflows is a research-preview feature that lets the model plan a hard problem and then fan it out across hundreds of parallel subagents. Each subagent does its share, the model verifies the work, and only then reports back. For example, Claude Code can now run a codebase-scale migration across hundreds of thousands of lines of code, from kickoff to merge, using the existing test suite as its pass-or-fail bar. The feature is available on Claude Code for Enterprise, Team and Max plans.
Users on claude.ai now control how much effort the model spends on a response. By default, Opus 4.8 runs at "high" effort, which uses roughly the same tokens as Opus 4.7 while delivering better results. For tougher jobs, you can push to "extra" or "max." Therefore, you decide the trade-off between speed, cost and quality on each task.
At ATF, we care about more than coding scores. The interesting story here is agentic financial analysis. Anthropic specifically calls out gains in this area, and the model now handles multi-step research, document reading and reasoning with fewer tool calls per task.
In plain terms, that means cheaper and more reliable AI analysts. An agent that reads filings, cross-checks figures and admits uncertainty is far more useful than one that confidently invents numbers. The honesty improvement directly reduces the risk of hallucinated data slipping into an investment thesis. For builders of AI-powered finance tools, this is the upgrade that actually moves the needle.
If you want to put these gains to work, our ATF Pro intelligence tools are built on exactly this kind of agentic pipeline. You can also read our breakdown of AI investing in 2026 for the bigger picture.
Anthropic claims Opus 4.8 beats both OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro on most key benchmarks. On agentic coding, for instance, Opus 4.8 scores 69.2% against 58.6% for GPT-5.5 and 54.2% for Gemini 3.1 Pro. It also leads on computer use and knowledge work.
Still, the picture is not a clean sweep. GPT-5.5 keeps a narrow edge on terminal-agent benchmarks. Meanwhile, lower-cost rivals such as DeepSeek and Grok win on raw price. So the right choice depends on your task: pick Opus 4.8 for coding, agents and finance work, and shop around when cost per token is the only thing that matters. If you want a cheaper model inside the Claude family, our complete Claude Sonnet 4.6 guide covers the lighter, faster option.
Migration is almost trivial. If you already build on Opus 4.7, the switch is essentially a one-line change.
claude-opus-4-8 (or claude-opus-4-8[1m] for the full 1M-token context).For most teams, that is the entire migration. On claude.ai and the Claude app, the new model and the effort control are simply available — no setup required.
Claude Opus 4.8 is a quality release dressed up as a minor version bump. The benchmark gains are real, the honesty improvement is genuinely useful, and the flat pricing makes the upgrade an easy decision. For investors, builders and analysts, the message is simple. You get a sharper, more trustworthy model for the same money — and a much cheaper fast mode on top. Anthropic has also hinted that its more powerful Claude Mythos models are coming in the next few weeks, so the pace is not slowing down.
Anthropic released Claude Opus 4.8 on May 28, 2026, just 41 days after Opus 4.7.
Standard pricing is $5 per 1M input tokens and $25 per 1M output tokens — the same as Opus 4.7. The fast mode costs $10 / $50 and runs at 2.5x the speed.
On Anthropic's benchmarks, Opus 4.8 leads GPT-5.5 on agentic coding, computer use and knowledge work. However, GPT-5.5 still holds a small lead on terminal-agent tasks.
The API model ID is claude-opus-4-8, or claude-opus-4-8[1m] for the 1M-token context window.
Sources: Anthropic's official Claude Opus page and launch announcement, May 28, 2026. Benchmark figures are self-reported by Anthropic.