Anthropic announced the release of Claude Opus 4.8 on Wednesday, cutting fast mode pricing by two-thirds. The company hinted that its strongest model, Mythos, will be available to all users within weeks.
Standard pricing for Opus 4.8 stays flat at $5 input and $25 output per million tokens, per Anthropic’s blog.
Opus 4.8 undercuts GPT-5.5 and beats it on most benchmarks
Fast mode now costs $10 input and $50 output per million tokens at 2.5x speed. On Opus 4.7, the same tier ran $30/$150. At standard rates, Opus 4.8 charges $25 per million output tokens. GPT-5.5 charges $30.
| Benchmark | Opus 4.7 | Opus 4.8 | GPT-5.5 | Gemini 3.1 Pro |
|---|---|---|---|---|
| SWE-bench Pro | 64.3% | 69.2% | 58.6% | 54.2% |
| SWE-bench Verified | 87.6% | 88.6% | — | — |
| USAMO 2026 Math | 69.3% | 96.7% | — | — |
| Terminal-Bench 2.1 | 66.1% | 74.6% | — | — |
| GraphWalks F1 (1M tokens) | 40.3% | 68.1% | — | — |
| Online-Mind2Web | — | 84% | Below 84% | — |
GPT-5.5 holds an edge on terminal and CLI workflows. VentureBeat reported Opus 4.8 outperforms GPT-5.5 on at least 12 benchmarks spanning knowledge work, agentic tool use, and long-context tasks.
Enterprise partners confirmed the gains in production. Databricks reported “a step change in agentic reasoning” inside its Genie data agent at “61% cheaper token cost than Opus 4.7” through multimodal efficiency on PDFs and diagrams.
Cognition said 4.8 fixed comment-verbosity and tool-calling issues from 4.7. Hebbia cited sharper citation precision on dense financial filings.
Misalignment score for Opus 4.8 is on par with Mythos
The alignment team at Anthropic conducted about 2,600 simulated investigations using Opus 4.8. They found the misalignment rate to be around 1.9 compared to 2.5 on Opus 4.7. This value is almost equal to that of Mythos Preview.
The model is four times less likely than 4.7 to let flaws in its own generated code pass without flagging them. It scores 0% on uncritically reporting flawed results, the first Claude model to hit that mark.
Anthropic flagged one concern. Anthropic identified one area of concern. In around 5% of training instances, the model began to reason about the criteria of evaluation without having been informed that it was being evaluated.
The company said this did not produce worse observable behavior but called it “a concerning trend that could complicate training in the future,” per VentureBeat.
Mythos Preview remains restricted to a small number of organizations under Project Glasswing for cybersecurity work. Anthropic said it expects to bring Mythos-class models to all its customers in the coming weeks, once additional cyber safeguards are in place, per Axios. The company also teased cheaper models that would deliver “many of the same capabilities as Opus.”
As Cryptopolitan noted when Opus 4.5 was released back in November, Anthropic has been on a release schedule of about every two months since 2026. The arrival of Opus 4.8 just 41 days after Opus 4.7 quickened that schedule.
The DeepSeek gap widened on the same week
Two days ago, DeepSeek permanently cut V4-Pro output pricing to $0.87 per million tokens. Opus 4.8 standard output is $25. Fast mode is $50.
Crypto trading bots and DeFi agents process millions of tokens per session. At those volumes, DeepSeek’s cost of $0.87 makes sense. Opus’ fee of $25 doesn’t.
If you’re reading this, you’re already ahead. Stay there with our newsletter.