GLM-4.7-Flash

(huggingface.co)

182 points | by scrlk 2 hours ago ago

48 comments

  • dajonker an hour ago ago

    Great, I've been experimenting with OpenCode and running local 30B-A3B models on llama.cpp (4 bit) on a 32 GB GPU so there's plenty of VRAM left for 128k context. So far Qwen3-coder gives the me best results. Nemotron 3 Nano is supposed to benchmark better but it doesn't really show for the kind of work I throw at it, mostly "write tests for this and that method which are not covered yet". Will give this a try once someone has quantized it in ~4 bit GGUF.

    Codex is notably higher quality but also has me waiting forever. Hopefully these small models get better and better, not just at benchmarks.

    • behnamoh 35 minutes ago ago

      > Codex is notably higher quality but also has me waiting forever.

      And while it usually leads to higher quality output, sometimes it doesn't, and I'm left with a bs AI slop that would have taken Opus just a couple of minutes to generate anyway.

    • latchkey an hour ago ago
      • WanderPanda 25 minutes ago ago

        I find it hard to trust post training quantizations. Why don't they run benchmarks to see the degradation in performance? It sketches me out because it should be the easiest thing to automatically run a suite of benchmarks

      • dajonker an hour ago ago

        Yes I usually run Unsloth models, however you are linking to the big model now (355B-A32B), which I can't run on my consumer hardware.

        The flash model in this thread is more than 10x smaller (30B).

  • vessenes 2 hours ago ago

    Looks like solid incremental improvements. The UI oneshot demos are a big improvement over 4.6. Open models continue to lag roughly a year on benchmarks; pretty exciting over the long term. As always, GLM is really big - 355B parameters with 31B active, so it’s a tough one to self-host. It’s a good candidate for a cerebras endpoint in my mind - getting sonnet 4.x (x<5) quality with ultra low latency seems appealing.

    • Workaccount2 25 minutes ago ago

      Unless one of the open model labs has a breakthrough, they will always lag. Their main trick is distilling the SOTA models.

      People talk about these models like they are "catching up", they don't see that they are just trailers hooked up to a truck, pulling them along.

    • HumanOstrich an hour ago ago

      I tried Cerebras with GLM-4.7 (not Flash) yesterday using paid API credits ($10). They have rate limits per-minute and it counts cached tokens against it so you'll get limited in the first few seconds of every minute, then you have to wait the rest of the minute. So they're "fast" at 1000 tok/sec - but not really for practical usage. You effectively get <50 tok/sec with rate limits and being penalized for cached tokens.

      They also charge full price for the same cached tokens on every request/response, so I burned through $4 for 1 relatively simple coding task - would've cost <$0.50 using GPT-5.2-Codex or any other model besides Opus and maybe Sonnet that supports caching. And it would've been much faster.

      • twalla 30 minutes ago ago

        I hope cerebras figures out a way to be worth the premium - seeing two pages of written content output in the literal blink of an eye is magical.

    • behnamoh 34 minutes ago ago

      > The UI oneshot demos are a big improvement over 4.6.

      This is a terrible "test" of model quality. All these models fail when your UI is out of distribution; Codex gets close but still fails.

    • ttoinou 28 minutes ago ago

      Sonnet was already very good a year ago, do open weights model right are as good ?

      • jasonjmcghee 24 minutes ago ago

        Fwiw Sonnet 4.5 is very far ahead of where sonnet was a year ago

    • mckirk 2 hours ago ago

      Note that this is the Flash variant, which is only 31B parameters in total.

      And yet, in terms of coding performance (at least as measured by SWE-Bench Verified), it seems to be roughly on par with o3/GPT-5 mini, which would be pretty impressive if it translated to real-world usage, for something you can realistically run at home.

  • montroser 21 minutes ago ago

    This is their blurb about the release:

        We’ve launched GLM-4.7-Flash, a lightweight and efficient model designed as the free-tier version of GLM-4.7, delivering strong performance across coding, reasoning, and generative tasks with low latency and high throughput.
    
        The update brings competitive coding capabilities at its scale, offering best-in-class general abilities in writing, translation, long-form content, role play, and aesthetic outputs for high-frequency and real-time use cases.
    
    https://docs.z.ai/release-notes/new-released
  • montroser 7 minutes ago ago

    > SWE-bench Verified 59.2

    This seems pretty darn good for a 30B model. That's significantly better than the full Qwen3-Coder 480B model at 55.4.

  • esafak 22 minutes ago ago

    When I want fast I reach for Gemini, or Cerebras: https://www.cerebras.ai/blog/glm-4-7

    GLM 4.7 is good enough to be a daily driver but it does frustrate me at times with poor instruction following.

  • bilsbie an hour ago ago

    What’s the significance of this for someone out of the loop?

    • epolanski an hour ago ago

      You can run gpt 5 mini level ai on your MacBook with 32 gb ram.

      You can get LLM as a service for cheaper.

      E.g. This model costs less than a tenth of Haiku 4.5.

  • baranmelik 40 minutes ago ago

    For anyone who’s already running this locally: what’s the simplest setup right now (tooling + quant format)? If you have a working command, would love to see it.

    • pixelmelt 14 minutes ago ago

      I would look into running a 4 bit quant using llama cpp (or any of its wrappers)

  • pixelmelt 15 minutes ago ago

    I'm glad they're still releasing models dispite going public

  • dfajgljsldkjag an hour ago ago

    Interesting they are releasing a tiny (30B) variant, unlike the 4.5-air distill which was 106B parameters. It must be competing with gpt mini and nano models, which personally I have found to be pretty weak. But this could be perfect for local LLM use cases.

    In my ime small tier models are good for simple tasks like translation and trivia answering, but are useless for anything more complex. 70B class and above is where models really start to shine.

  • karmakaze 2 hours ago ago

    Not much info than being a 31B model. Here's info on GLM-4.7[0] in general.

    I suppose Flash is merely a distillation of that. Filed under mildly interesting for now.

    [0] https://z.ai/blog/glm-4.7

    • lordofgibbons an hour ago ago

      How interesting it is depends purely on your use-case. For me this is the perfect size for running fine-tuning experiments.

    • redrove an hour ago ago

      A3.9B MoE apparently

  • eurekin an hour ago ago

    I'm trying to run it, but getting odd errors. Has anybody managed to run it locally and can share the command?

  • twelvechess 2 hours ago ago

    Excited to test this out. We need a SOTA 8B model bad though!

  • XCSme 2 hours ago ago

    Seems to be marginally better than gpt-20b, but this is 30b?

    • strangescript 2 hours ago ago

      I find gpt-oss 20b very benchmaxxed and as soon as a solution isn't clear it will hallucinate.

      • blurbleblurble an hour ago ago

        Every time I've tried to actually use gpt-oss 20b it's just gotten stuck in weird feedback loops reminiscent of the time when HAL got shut down back in the year 2001. And these are very simple tests e.g. I try and get it to check today's date from the time tool to get more recent search results from the arxiv tool.

    • lostmsu 2 hours ago ago

      It actually seems worse. gpt-20b is only 11 GB because it is prequantized in mxfp4. GLM-4.7-Flash is 62 GB. In that sense GLM is closer to and actually is slightly larger than gpt-120b which is 59 GB.

      Also, according to the gpt-oss model card 20b is 60.7 (GLM claims they got 34 for that model) and 120b is 62.7 on SWE-Bench Verified vs GLM reports 59.7

  • epolanski 2 hours ago ago

    Any cloud vendor offering this model? I would like to try it.

    • PhilippGille 2 hours ago ago

      z.ai itself, or Novita fow now, but others will follow soon probably

      https://openrouter.ai/z-ai/glm-4.7-flash/providers

      • epolanski 2 hours ago ago

        Interesting, it costs less than a tenth than Haiku.

        • saratogacx an hour ago ago

          GLM itself is quite inexpensive. A year sub to their coding plan is only $29 and works with a bunch of various tools. I use it heavily as a "I don't want to spend my anthropic credits" day-to-day model (mostly using Crush)

    • dvs13 2 hours ago ago
    • latchkey an hour ago ago

      We don't have lot of GPUs available right now, but it is not crazy hard to get it running on our MI300x. Depending on your quant, you probably want a 4x.

      ssh admin.hotaisle.app

      Yes, this should be made easier to just get a VM with it pre-installed. Working on that.

      • omneity an hour ago ago

        Unless using docker, if vllm is not provided and built against ROCm dependencies it’s going to be time consuming.

        It took me quite some time to figure the magic combination of versions and commits, and to build each dependency successfully to run on an MI325x.

        • latchkey an hour ago ago

          Agreed, the OOB experience kind of suck.

          Here is the magic (assuming a 4x)...

            docker run -it --rm \
            --pull=always \
            --ipc=host \
            --network=host \
            --privileged \
            --cap-add=CAP_SYS_ADMIN \
            --device=/dev/kfd \
            --device=/dev/dri \
            --device=/dev/mem \
            --group-add render \
            --cap-add=SYS_PTRACE \
            --security-opt seccomp=unconfined \
            -v /home/hotaisle:/mnt/data \
            -v /root/.cache:/mnt/model \
            rocm/vllm-dev:nightly
            
            mv /root/.cache /root/.cache.foo
            ln -s /mnt/model /root/.cache
            
            VLLM_ROCM_USE_AITER=1 vllm serve zai-org/GLM-4.7-FP8 \
            --tensor-parallel-size 4 \
            --kv-cache-dtype fp8 \
            --quantization fp8 \
            --enable-auto-tool-choice \
            --tool-call-parser glm47 \
            --reasoning-parser glm45 \
            --load-format fastsafetensors \
            --enable-expert-parallel \
            --allowed-local-media-path / \
            --speculative-config.method mtp \
            --speculative-config.num_speculative_tokens 1 \
            --mm-encoder-tp-mode data
    • xena 2 hours ago ago

      The model literally came out less than a couple hours ago, it's going to take people a while in order to tool it for their inference platforms.

      • idiliv 2 hours ago ago

        Sometimes model developers coordinate with inference platforms to time releases in sync.