Just 41 days after deploying its previous AI model update, Anthropic officially launched Claude Opus 4.8. This unusually rapid release cycle follows a lukewarm reception to the previous version and intense competitive pressure from recent upgrades to OpenAI’s Codex and Google’s Gemini Flash platform.
Claude Opus 4.8 arrives with a rare feature for an AI: Honesty
The headline improvements for Opus 4.8 center around software engineering capabilities and overall accuracy. In standard evaluations, the model achieved a 69.2% score on the SWE-Bench Pro test. This successfully outperformed rival industry giants such as GPT-5.5 and Gemini 3.1 Pro. OpenAI still retains a lead on terminal-coding-specific tasks, though.
Beyond raw programming prowess, Anthropic focused heavily on fixing structural data errors. According to early testers, the new software demonstrates a dramatic surge in honesty. The model is roughly four times less likely to allow flawed code to pass by unremarked compared to its predecessor. Instead of hallucinating answers when dealing with messy or uncertain inputs, Opus 4.8 actively flags its own uncertainties to the user. Investment giant Bridgewater Associates noted that the biggest difference in this upgrade is the model’s unique tendency to proactively catch data flaws that other platforms routinely miss.
Parallel subagents and effort control
Alongside the core intelligence upgrade, Anthropic is introducing a research preview tool called Dynamic Workflows. This system allows the platform to organize and run hundreds of parallel subagents within a single session. Using this feature inside Claude Code, the platform can handle massive, codebase-scale system migrations stretching across hundreds of thousands of lines of code.
Users also gain access to a new “Effort Control” toggle. This feature lets you choose exactly how much computational power the system puts into a specific response. Selecting a lower setting triggers faster replies while consuming your daily rate limits much more slowly. Opus 4.8 defaults to high effort to maintain the highest quality output. Additionally, developers using the Messages API can now input system instructions mid-task to adjust the model’s behavior on the fly.
Pricing and availability
Opus 4.8 is available globally starting today. Despite the performance gains and a fast mode that runs 2.5 times quicker, standard pricing remains identical to the prior iteration.
The post Anthropic Drops Claude Opus 4.8 to Reclaim the Generative AI Coding Crown appeared first on Android Headlines.
