ReBlink Introduces AI-Driven Gameplay, powered by Groq

ReBlink, the game studio behind ARBO Arena Tactics, is using AI to redefine how players interact with strategy games. Founded by Arunan Sri, the company’s mission is to make deep, tactical gameplay not only fun, but accessible to a wider audience, including casual players or those with motor skill limitations like Parkinson’s. Arunan and his team set out to answer a bold question: What if the most complex part of a strategy game wasn’t understanding the controls or mechanics, but the thinking? And what if the game could share some of that thinking with you?

Their answer is Arena Tactics, a next-generation tactics game with collectible card mechanics where players now have an alternate game mode to just clicking through radial menus or memorizing hotkeys. Instead, they command their armies by speaking or typing natural language. A player can write a full tactical plan (“flank right, then push mid once shields are up”), or toss out quick commands in the heat of battle (“move north,” “base rush,” “attack the unit with lower health”).

ReBlink isn’t trying to build just another strategy game, though. They’re transforming how people interact with the strategy itself. “At the heart of Arena Tactics is a simple idea: language is a control mechanic,” Arunan says. “Over 800 million people use ChatGPT every week. The world is learning how to prompt. If people already use language to express intent, why shouldn’t they use it to play games with deeper mechanics?”

Making deep strategy fast and accessible

ReBlink wanted to bring deep, expressive tactics to every kind of player. But the very complexity that makes Arena Tactics exciting also introduced a serious technical hurdle. The AI needed to understand spatial relationships, plan moves, and manage layers of card combinations, all in real time.

Early prototypes built on custom models, Llama 3, and Gemini lacked the speed and spatial reasoning needed to keep players engaged. There wasn’t enough compelling single-player content to keep people playing. In Prompt Battler mode, which is fully controlled by language, players struggled to get the AI to behave as intended, and slow inference from previous cloud providers made matches feel long and draggy, so much so that an investor flagged pacing as a major issue. As a result, Prompt Battler made up less than 5% of total games, Incursions, which is single-player manual control campaign mode versus an AI powered by this architecture, were under 10%, and even Tactical Mode, the traditional core strategy gameplay, against AI saw poor repeat play due to the lack of dynamic solo experiences.

ReBlink needed an AI backend that could think as fast as players do, and that’s where Groq comes in.

The breakthrough: Groq powers real-time AI strategy

The team rebuilt the AI engine from the ground up using a multi-agent AI framework with LangChain and LangGraph, then plugged in GroqCloud to handle all LLM-driven reasoning. That’s what fuels the Planning Agent, the game’s “brain”, which interprets commands, reads the battlefield, and runs constant simulations to choose the best moves.

Then, when ReBlink adopted the GPT-OSS-20B model running on GroqCloud, everything changed. “We compared Groq running GPT-OSS-20B vs Google's Gemini 2.5 Flash model and OpenAI's GPT-5-nano,” Arunan explains. And the results speak for themselves:

• Per-call latency: Average model response time dropped from 14.35 seconds on Gemini and 26.19 seconds on GPT-5-nano to 3.82 seconds on Groq. This is roughly 4x lower latency than Gemini and ~7x lower than OpenAI, which is what enables much more fluid, near real-time command execution in-game.
• End-to-end gameplay time: Total model time per full game went from 770.52 seconds on Gemini and 1580.05 seconds on GPT-5-nano down to 184.56 seconds on Groq. This means Groq runs a full match ~ 4x faster than Gemini and ~8.5x faster than the OpenAI baseline, directly improving pacing and responsiveness for players.
• Cost efficiency: The cost per game decreased about 14x less LLM spend than Gemini and ~2.3x less than OpenAI. Additionally, on a per-round basis, Groq is roughly 2x cheaper than OpenAI and ~15x cheaper than Gemini.

“After introducing our multi-agent AI framework with GroqCloud, everything shifted. Tactical Mode, Incursions, and Prompt Battler jumped to over 60% of total plays,” Arunan shares. “This means that over 60% of all games were played in modes that relied directly on our AI architecture running on GroqCloud, whether players were using the fully language controlled mode or manual Tactical Mode against an AI.”

“Groq gives us both significantly lower latency and order-of-magnitude lower cost versus the other providers, which is why we're leaning on it for real-time gameplay experiences,” he continued. “Groq’s track record on performance and cost gives us the confidence to continue underwriting our research and development of novel AI co-play and language control mechanics.”

A new type of co-play: Strategy with a tactical companion

What emerged from this redesign isn’t just faster gameplay, it’s a new relationship between the player and AI. “The AI acts like a tactical companion,” Arunan explains. “It learns from the deckbuilding and context the player sets, then plans and executes intelligently. If the player gives a clear command, it follows. If the command has gaps, the AI fills them based on its own reasoning.”

The result is a kind of strategic back-and-forth, occasionally even with moments of playful disagreement, a feeling that the game is less a system to control and more a teammate to collaborate with.

This co-play dynamic is only possible because Groq gives the Planning Agent near-instant reasoning and structured responses. Instead of a static script, the AI adapts, clarifies intent, and helps players discover strategies they may not have considered.

The ripple effects: Accessibility, cost efficiency, and fun for all

“Getting here wasn’t as simple as plugging in an LLM API,” Arunan says. “We’ve been exploring this space since the fall of 2022, figuring out how to combine AI co-play with strategic depth and high visual fidelity. We believe we’re the first to bring something of this depth, polish, and AI integration to market in a way that feels like an actual game, not just a tech demo.”

“When latency stops getting in the way, commands feel instantaneous. The AI can evaluate thousands of possibilities per second, giving Arena Tactics a responsiveness that feels immediate and real, but in a way that serves the player and doesn’t overwhelm them,” says Arunan.

Building a new genre: Language as the next game controller

ReBlink is pioneering a new genre, where language is a control mechanic for games. The team believes Arena Tactics represents more than a product milestone; it’s the start of a shift in how deep mechanic games are played and enjoyed.

Groq makes it possible for ReBlink to maintain this high-speed experience across PC, Mac, and mobile, with a 2026 launch in the works that will let players control a deep strategy game with natural language, for a market first.

And they see this as just the beginning. “As AI evolves, it will shift from being something in the background to becoming an active collaborator. It won’t just respond. It will challenge, adapt, and strategize with players. That’s the future we’re building toward,” Arunan says.

Powered by Groq, ReBlink is ushering in a new era of gaming that is broadly accessible, and deeply human. “Speed is everything,” Arunan emphasizes. “Inference is the fuel that powers that entire matrix of options. Groq gives us that fuel.”