Performance remains largely unknown, but according to TFLOPS calculations we've seen, it could be close to a GeForce RTX 3080.
TFlops calculations are a terrible way to estimate graphics card performance for cards using a new architecture, especially for a brand new line of cards with no direct predecessor. Even going from Nvidia's 20 series to their current 30 series, there was a huge difference in how much gaming performance a given number of TFlops works out to. Going by advertised compute performance alone, one would expect a "20.3 Tflop" 3070 to be around 50% faster than a "13.5 TFlop" 2080 Ti, but in reality, the two typically perform quite similar in games, since the gaming-performance relative to compute performance was much lower for the 30-series. It's also hard to base any comparisons on the ultra-low-end Xe graphics hardware that's already been released, since it's unknown how well the architecture will scale to larger sizes with faster memory.
Most rumors I've seen have suggested that their top-end part might get around 3070 performance or so. Without knowing more details, it's possible that it could manage to be more of a 3080 competitor, but a sub-heading that suggests "Potential RTX 3080 performance for around $650–$850" amounts to little more than complete speculation. And even the approximate value of the prizes doesn't tell us that much when they don't go into any detail about what the "Intel Arc branded merchandise" will include. At best, we can consider these upper limits for the value of each card.
And as has been pointed out, suggested pricing doesn't mean much for the street prices of cards right now, and it's hard to say exactly how long it might take before pricing for a given level of performance gets back to where it was a year ago.