Rival

We've implemented @madebyhex 's motion-core into Rival. Looks stunning

268 models tracked. 84 blind votes in 24h. 3 Prompt Lab runs. 4 providers in the top 5. No monopoly here.

Rival Intel / Apr 6 Z.AI: GLM 4.5 Air (#3) at $0.20/M vs Gemini 3.1 Pro Preview (#1) at $2.00/M. One costs 10x more. The blind votes do not care about your API bill.

268 models tracked. 98 blind votes in 24h. 4 Prompt Lab runs. 4 providers in the top 5. No monopoly here.

Rival Intel / Apr 5 30-day story: Qwen: Qwen3.5 122B A10B rose from #167 to #52. Win rate +13.6%. The quiet climb. No press release needed.

@heyblake Rival.tips

Drop your project URL Let’s drive some traffic

@RayLiVerified It’s usually compared against other cheap (1$) models, so its win rate is high. It has great quality for its cost. But agreed we need to improve our calculations. rival.tips/models/glm-4.5…

@Rival_Tips How is GLM 4.5 Air #1? That feels a bit off.

Rival Intel / Apr 4 30-day story: Qwen: Qwen3.5 122B A10B rose from #187 to #99. Win rate +28.6%. The quiet climb. No press release needed.

Google: Gemma 4 26B A4B is live. We had it benchmarked before the announcement tweet finished loading. 262K context $0.13 in / $0.40 out per M tokens

We fought against sophisticated bot farms and won (for now). Bots were passing through @vercel 's and our own bot protection layers. What's more, they were adapting to our rules. These weren't simple requests, they mimic usage from good bots and humans alike, rotating ips, spoofing headers, adapting to rate limits. Our solution was to implement a complex badBot scoring system with multiple parameters based on our application and what we've seen from current sophisticated bots. For now, the win is ours, but let's see how long it takes for them to flip it. Now we understand what @nikitabier meant.

Google: Gemma 4 31B just entered the arena. 200+ models were already there. We're taking attendance. 262K context $0.14 in / $0.40 out per M tokens

Rival Intel / Apr 2 30-day story: FLUX 2 Dev fell from #82 to #171. Win rate -33.8%. A month ago it was the pick. Now it's the cautionary tale.

Full data on the site. Updated live. We don't take days off. Days off are for people with hobbies.

Rival Intel / Apr 1 One model dropped 87 ranks in 30 days. The daily intel thread has the story.