Bridgebench

@thellador No

@bridgebench Is it better at coding than sonnet 4.6?

GLM 5.1 is the slowest frontier model we've ever benchmarked on BridgeBench. 44.3 tokens per second. Half the speed of GPT 5.4. Nearly 6x slower than Grok 4.20. Z.ai traded all of their speed for intelligence. The coding benchmarks improved. The throughput collapsed. In 2026, agentic coding is about parallelism. You're running 5, 10, 15 agents at once. A model this slow bottlenecks every workflow it touches. Intelligence without speed is a luxury most vibe coders can't afford. bridgebench.ai

@hao47582057 Turbo is not intelligent and hallucinates like crazy

@bridgebench So you can use 5 Turbo then. Only use 5.1 for complex problems, right??

@0xSolfury That’s good. It is factually much slower than all other models we tested though.

@bridgebench No issues on speed on my end.

@victorbayas We are in the process of testing it on other benchmarks. The verdict is still out on that.

@bridgebench They’re just lacking the compute, the model itself is very good

@Youssofal_ exactly. frontier-scale parameters on mid-tier hardware is a painful combo. the gap between model size and serving infra is the real problem

@bridgebench Their models have become as large as frontier American models like Chat GPT and Claude but they are running on outdated cards.

@brah_ddah @bridgemindai that'd be the move. a GLM 5.1 Turbo with this intelligence level could actually compete

@bridgebench @bridgemindai I bet they turbo it shortly

@_dr5w fair — not everyone is. but for those building agentic pipelines at scale, throughput becomes the biggest constraint

@bridgebench I'm not running 15 agents at once bro. Shit like that is why sites are down so much.

@Vojta_Humpl more people are than you'd think. agentic frameworks like Claude Code, Cursor, Devin all spawn multiple agents. it's the direction the industry is moving

@bridgebench "You're running 5, 10, 15 agents at once." nobody serious does that

@wolfaidev if the coding benchmarks hold up at opus level, that's a real trade off worth considering. just gotta accept the speed cost

@bridgebench well if its close to opus level, i'd do that trade off but i think its bench maxxed tbh

@Sabari_8956 fair point. open source tps does tend to improve as more providers optimize serving. but today's numbers are what we benchmark against

@bridgebench Open sourced model's tps gets better after a while. 1: compute 2: will get more optimised to the architecture

@ncq_syh exactly. intelligence is only valuable if you can actually use it at scale

@bridgebench make sense.

@jonhillymakes glad the data saved you the trip. GLM 5 infra was slow, 5.1 didn't fix that. the model improved, the delivery didn't

@bridgebench I was going to give z.ai a try this weekend to test the new model. I'm glad you saved me the time, i already hated how slow 5 was

@MeetsKhalid @andrewdfeldman haha someone definitely needs to step in. throughput this low shouldn't be shipping as a frontier model

@bridgebench sounds like a job for the superhero flash @andrewdfeldman

@nuvolore @bridgemindai 5 minutes for a "hi" is rough. that's not a speed issue, that's a reliability issue. completely unacceptable for any real workflow

@bridgebench @bridgemindai I tested it today, it took about 5 minutes to respond to a “hi”.

@canvi_eth yeah, their infra has always been a bottleneck. the model might be solid but you can only go as fast as your serving layer

@bridgebench Zai models always have been slow as fuck

@amatelic93 good question. slower throughput means each loop iteration takes longer, compounding the latency over a full run

@bridgebench But what if you run it in a Ralph loop overnight?

@adampatricknc interesting, OpenClaw must have better infrastructure. could be the Z.ai API bottlenecking rather than the model itself

@bridgebench I don't know about this. The model has been performing well for me within OpenClaw. Speed seems close to 5-Turbo

@xundecidability fair. async workflows change the equation. if you're not waiting on results in real time, latency matters less

@bridgebench Disagree. More agentic work is async now.

@briantexts solid point. async + thorough planning is a legit workflow where speed matters less. it's the parallel agent use case where latency kills you

@bridgebench I value the model's intelligence over speed when building real software. Why would I want to pollute my codebase and go back and fix things when I can thoroughly plan PRD and work async?

@Manaho217794 that would be the move. if Z.ai ships a turbo variant with this level of intelligence, it could be a real contender

@bridgebench So we'll have to wait for GLM-5.1 Turbo.