Capcom is turning RE Engine into an 'AI-generation game engine'
A Japanese gaming giant just committed its flagship engine to AI-native development, and a Chinese Flash-class model is putting up frontend work that holds up next to Opus 5.5.
Capcom has laid out a plan to gradually transform RE Engine into an "AI-generation game engine." At the Capcom Open Conference RE: 2026 on October 2, programmer Satoshi Ishida presented the REX project — a multi-step overhaul of the engine behind Resident Evil and Street Fighter — arguing that "successfully integrating AI technology into development workflows has become an essential condition for streamlining future development." The framing matters: Capcom is not generating art assets with AI. Its line, repeated at a March 2026 shareholders meeting, is that AI-generated assets will not ship in its games, but the technology will be used aggressively behind the scenes — in graphics, sound, and programming — to cut the cost of making enormous games. REX unifies the engine under what GameBiz describes as the world's most common programming language, making it legible to both human developers and AI tools, and Capcom plans to open-source at least some of the REX components specifically so AI models can be trained on them. This is one of the first times a AAA publisher has put a named engine roadmap and a public "AI-generation" target behind what has so far been mostly hedged language, and it turns the industry's loudest AI debate — assistance versus replacement — into something concrete you'll be able to evaluate when the next Resident Evil ships.
MiniMax's M3.1-Flash is producing frontend work that Chinese reviewers say holds up against Claude Opus 5.5. The model, which entered public beta on the eve of China's National Day, was put through a series of creative frontend tasks by Zhidx — procedural animation, an interactive "starlight train" scene, a cocktail-making explainer, and a Canvas pixel-wizard — and the outlet's verdict was that several effects it assumed required a flagship model came out of a Flash-tier one, with cleaner step breakdowns than Opus 5.5 in at least one test and its own visual embellishments in another. MiniMax attributes this to the M3 series' native multimodal training and 1M-token context window; the same team's open-source H3 model reportedly passed 24 million downloads and 300 public derivatives within a month. The claim to watch is really about economics: if Flash-tier pricing buys near-flagship frontend output, the premium for frontier models narrows on exactly the task category — working UI — that application builders pay it for.
What to watch: whether Capcom's "AI-generation engine" language survives contact with developers at this month's reveal cycles, and whether M3.1-Flash's frontend showing gets reproduced outside Chinese review outlets.
Where do you come down on Capcom's line — AI in the pipeline but never in the shipped asset? Tell us in the comments.
