Alex Getmantidlige AI-signaler
DA

GLM-4.7 Flash

Z.ai

LanceretOpdateret
Kontekstvindue
203K
202.752 tokens

Tidslinje

  1. · Vægtversion · kildelink
  1. · Kildelinks · GitHub search · Z.ai · pulls · ivanmlnnn/sentinel · @ivanmlnnn
    Uddrag
    binding `AI`, shared `functions/api/_ai.js` adapter, and configured model `@cf/zai-org/glm-4.7-flash`.
  2. · Kildelinks · GitHub search · Z.ai · commits · Nitjsefnie/ai-researcher · @github-actions[bot]
    Uddrag
    GLM-4.7-Flash (Reasoning) [Z AI]
  3. · Kildelinks · GitHub search · Z.ai · pulls · sourceplane/pdf-quiz-generator · @pullely
    Uddrag
    Workers AI through createAI (agents/models/ai-sdk) over the binding, @cf/zai-org/glm-4.7-flash, or Gemini 2.5 Pro and Flash when GOOGLE_GENERATIVE_AI_API_KEY is set — and its schemaInPromptMiddleware moves a
  4. · Kildelinks · GitHub search · Z.ai · pulls · sourceplane/pdf-quiz-generator · @pullely
    Uddrag
    the Worker's binding, env.AI.toMarkdown for the PDF and @cf/zai-org/glm-4.7-flash for the quiz and title, through createAI from agents/models/ai-sdk, which replaced the deprecated workers-ai-provider and needs
  5. · Kildelinks · GitHub search · Z.ai · pulls · sgl-project/sglang · @1195343015
    Uddrag
    --model-path ZhipuAI/GLM-4.7-Flash \
  6. · Kildelinks · GitHub search · Z.ai · pulls · praveens6050-afk/Fashion_Fussion · @praveens6050-afk
    Uddrag
    - Preview uses @cf/zai-org/glm-4.7-flash, selected for Workers Free compatibility.
  7. · Kildelinks · GitHub search · Z.ai · pulls · luongnv89/freetokens · @luongnv89
    Uddrag
    | bumped | — | — | \| GLM-4.7-Flash \| Free \| Free \| Free \| Free \| |
  8. · Kildelinks · GitHub search · Z.ai · pulls · diazMelgarejo/Perpetua-Tools · @diazMelgarejo
    Uddrag
    `_resolve_candidates()`. Isolated reproduction on the pre-fix tree: both `glm-4.7-flash` and `openrouter/free` returned ready while the qualification gate denied them, and the probe-failure fallback (`resolved or
  9. · Kildelinks · GitHub search · Z.ai · commits · Nitjsefnie/ai-researcher · @github-actions[bot]
    Uddrag
    GLM-4.7-Flash (Reasoning) [Z AI]
  10. · Kildelinks · GitHub search · Z.ai · pulls · wwoosshh/GeulPUv2 · @wwoosshh
    Uddrag
    (1,299 / 30,076 rows) plus 8 greedy tokens. GLM-4.7-Flash · Kanana-2 30B-A3B
  11. · Kildelinks · GitHub search · Z.ai · pulls · Sys64k/ai-swarm-orchestrator · @Sys64k
    Uddrag
    its weights fit in 16 GB. qwen3-coder and glm-4.7-flash don't, so part of each ran on the CPU. That makes them slower but shouldn't change
  12. · Kildelinks · GitHub search · Z.ai · commits · logicplanes/kaeso · @dvnold
    Uddrag
    `@cf/zai-org/glm-4.7-flash`, still gets "is a generation model".
  13. · Kildelinks · GitHub search · Z.ai · pulls · paracha34/zara · @paracha34
    Uddrag
    Gemini chat path with Cloudflare Workers AI model `@cf/zai-org/glm-4.7-flash`.
  14. · Kildelinks · GitHub search · Z.ai · pulls · fumitakaojima/ai-frontier-news · @fumitakaojima
    Uddrag
    Cloudflareモデル仕様：https://developers.cloudflare.com/workers-ai/models/glm-4.7-flash/
  15. · Kildelinks · GitHub search · Z.ai · pulls · Sys64k/hardwarestack · @Sys64k
    Uddrag
    - **Calculator:** adds Qwen3-Coder 30B-A3B, GLM-4.7-Flash 30B-A3B (new `GLM` family, MLA KV cache like DeepSeek), Devstral Small 24B (2507) and Qwen2.5-Coder 14B. Architecture is from
  16. · Kildelinks · GitHub search · Z.ai · pulls · Sys64k/ai-swarm-orchestrator · @Sys64k
    Uddrag
    - **Correction:** the page said only qwen3-coder and glm-4.7-flash ran partly on the CPU. Every model did at the test's context. The column is renamed
  17. · Kildelinks · GitHub search · Z.ai · pulls · NousResearch/hermes-agent · @DevvGwardo
    Uddrag
    sent to `api.surplusintelligence.ai` (model `glm-4.7-flash`) to compress
  18. · Kildelinks · GitHub search · Z.ai · pulls · Cap-go/Capve · @TorichanCapgo
    Uddrag
    Adcash zone bid-change recommendations from Cloudflare Workers AI (GLM-4.7-Flash). Every existing and new campaign remains disabled by default.
  19. · Kildelinks · GitHub search · Z.ai · commits · PowerTokens/video-studio · @PowerTokens
    Uddrag
    catalog. The pickers now share four models, with glm-4.7-flash still the free default. A saved selection of the removed model falls back to that default.
  20. · Kildelinks · GitHub search · Z.ai · commits · Ogires/horoscopo-de-bardo · @Ogires
    Uddrag
    En la prueba real, GLM-4.7-Flash con thinking gastó los 4.096 tokens de
  21. · Kildelinks · GitHub search · Z.ai · pulls · vllm-project/vllm · @mgoin
    Uddrag
    `ops.lut_b_moe_gemm` timing, GLM-4.7-Flash expert shapes, cold L2 (1 GiB memset before each launch), CUDA events, median of 30, µs, before → after. The instruction is
  22. · Kildelinks · GitHub search · Z.ai · pulls · rasikur077-arch/lili-bilingual-companion · @rasikur077-arch
    Uddrag
    - Use Workers AI model `@cf/zai-org/glm-4.7-flash` when the binding is present.
  23. · Kildelinks · GitHub search · Z.ai · pulls · logicplanes/kaeso · @dvnold
    Uddrag
    Chat Completions request and stream through Workers AI (`@cf/zai-org/glm-4.7-flash`), costs and `cf-aig-log-id` in the request log, and the same requests visible in Cloudflare's own gateway
  24. · Kildelinks · GitHub search · Z.ai · pulls · kennedy-blip/Julesown · @kennedy-blip
  25. · Kildelinks · GitHub search · Z.ai · pulls · vllm-project/vllm · @mgoin
    Uddrag
    - GSM8K, `zai-org/GLM-4.7-Flash`, full 1319 questions, 5-shot chat, thinking off, max 2048 generated tokens, `--quantization lut_b_moe` on the native SM107 path (`tcgen05.mma.kind::mxf8f6f4.block_scale.decompress::lut::b.block32`, MXFP8 activations):
  26. · Kildelinks · GitHub search · Z.ai · pulls · marromugi/iris · @marromugi
    Uddrag
    `Model: <名前>.` が付く。`MODEL_PROVIDER` が `anthropic` なら `claude-haiku-5-5`、`workers-ai` なら `@cf/zai-org/glm-4.7-flash`。
  27. · Kildelinks · GitHub search · Z.ai · pulls · judecutler/wugs · @judecutler
    Uddrag
    4 sections by BM25, and asks Workers AI `@cf/zai-org/glm-4.7-flash` (Cloudflare's replacement for the Llama 3.x models it retires on Nov 20) to answer from those
  28. · Kildelinks · GitHub search · Z.ai · pulls · hireymage/voice-agent-mcp · @hireymage
    Uddrag
    Tried on a device: ending by voice with GLM-4.7-Flash ("Ukonči se", "Zatím se měj", "Vypni se" ended 2/2; "Počkej chvilku" 0/2). Whether a "be quiet"
  29. · Kildelinks · GitHub search · Z.ai · pulls · AndresYepez23/PRISMA · @AndresYepez23
    Uddrag
    - Assert the official general Z.AI endpoint and GLM-4.7-Flash model remain configured.
  30. · Kildelinks · GitHub search · Z.ai · pulls · 1bit-MONSTER/llama.cpp · @bong-water-water-bong
    Uddrag
    - GLM-4.7-Flash Q4_K_M, ~750-token prompt through the prefill kernel with #70's row-major V, 4 runs with a different request in between: bit-identical logits. Before: up
  31. · Kildelinks · GitHub search · Z.ai · pulls · 1bit-MONSTER/llama.cpp · @bong-water-water-bong
    Uddrag
    ggml-hrx: no V transpose for MLA strided V (GLM-4.7-Flash prompts >= 512 tokens: PPL 315,664 -> 5.92)
  32. · Kildelinks · GitHub search · Z.ai · pulls · 1StepMore/Omni_Localizer · @1StepMore
  33. · Kildelinks · GitHub search · Z.ai · pulls · luongnv89/freetokens · @luongnv89
  34. · Kildelinks · GitHub search · Z.ai · pulls · ai-dynamo/dynamo · @ishandhanani
    Uddrag
    own Modal node: two `dynamo.sglang` TP2 workers of GLM-4.7-Flash on 4x H200 (1,767,648 KV tokens per worker, 3.5M total), `dynamo.frontend --router-mode kv --router-session-affinity-mode soft`, one
  35. · Kildelinks · GitHub search · Z.ai · pulls · ROCm/aiter · @hyjuunn
    Uddrag
    backend calling this AITER kernel showed 1.17x faster GLM-4.7-Flash TP4 decode steps versus AITER `16979d757` (geomean of 10 workload shapes). This is not an upstream-vLLM
  36. · Kildelinks · GitHub search · Z.ai · pulls · logicplanes/kaeso · @dvnold
    Uddrag
    for them. A generation model there, such as `@cf/zai-org/glm-4.7-flash`, still gets "is a generation model".
  37. · Kildelinks · GitHub search · Z.ai · pulls · JingbiaoMei/Tokdash · @JingbiaoMei
    Uddrag
    | `glm-4.7-flash` | price_change | additions-only automation: existing model prices require a manual update |
  38. · Kildelinks · GitHub search · Z.ai · pulls · cheaper-inference/OmniRoute · @kainatquaderee
    Uddrag
    doubled closer (D10), on three different models** (`deepseek-v4-flash-0731`, `glm-4.7-flash`, `llama-4-scout-17b-16e-instruct`): `<write>{…}</path>` and `<write>{…}</write></path>` both come back as `finish_reason: "tool_calls"`, `write`, with the arguments intact and
  39. · Kildelinks · GitHub search · Z.ai · pulls · B-E-L-I-E-V-E/Personal-AI-Router-Plus · @B-E-L-I-E-V-E
    Uddrag
    - [ ] Live: **Apply now** on the GLM-4.7-Flash plan for Ultron (131,072 × 5). Needs this build on Ultron and the user's OK to
  40. · Kildelinks · GitHub search · Z.ai · commits · Nitjsefnie/ai-researcher · @github-actions[bot]
    Uddrag
    GLM-4.7-Flash (Non-reasoning) [Z AI]
  41. · Kildelinks · GitHub search · Z.ai · pulls · sgl-project/sglang · @divakantpandey
    Uddrag
    only, no model forward changes. GPU check on GLM-4.7-Flash (before/after, logs) is in a comment below.
  42. · Kildelinks · GitHub search · Z.ai · pulls · rolexdkscreation-max/DK2.0 · @rolexdkscreation-max
    Uddrag
    - नए providers: `zai` (https://api.z.ai/api/paas/v4, default glm-4.7-flash) और `nvidia` (https://integrate.api.nvidia.com/v1, default meta/llama-3.3-70b-instruct)।
  43. · Kildelinks · GitHub search · Z.ai · pulls · leehack/llamadart · @leehack
    Uddrag
    contentOnly, and for the structured-output route with the GLM-4.7-Flash template (`<think>Plan.</think>{"a":1}` gives reasoning `<think>Plan.`, as the oracle) (`chat_format_registry_test.dart`).
  44. · Kildelinks · GitHub search · Z.ai · pulls · antonellof/frink · @antonellof
    Uddrag
    and `drop_thinking`. frink set none of them, so GLM-4.7-Flash, LFM3, LFM2.5 and Nemotron-3-Nano always dropped earlier turns' reasoning, Laguna-S-2.1 did with thinking off, and DeepSeek-V4
  45. · Kildelinks · GitHub search · Z.ai · pulls · NovaSky-AI/SkyRL · @ArturNiederfahrenhorst
    Uddrag
    suite on 4×L4, and the Kimi K2.5 and GLM-4.7-Flash rows on 4×H100. Megatron-only; FSDP does not use Megatron-Bridge.
  46. · Kildelinks · GitHub search · Z.ai · pulls · sgl-project/sglang · @HollowMan6
    Uddrag
    - `glm4_moe_lite.py` (GLM-4.7-Flash and its NextN draft): drop the three `q/k/v_proj -> qkv_proj` entries from `stacked_params_mapping`, which the MLA attention has no parameter for and
  47. · Kildelinks · GitHub search · Z.ai · pulls · kahnwong/self-hosted · @renovate[bot]
    Uddrag
    Add PowerTokens (`api.powertokens.ai`) as an OpenAI-compatible provider. Its `glm-4.7-flash` model is listed at $0 (0 Credits); chat, streaming and tool calling were tested through the
  48. · Kildelinks · GitHub search · Z.ai · pulls · Danish797035/harness-agent · @Danish797035
    Uddrag
    Gemini 2.5 Flash, Groq Llama 3.3 70B, Z.ai GLM-4.7-Flash, Gemini 3.1 Flash Lite; Kimi K2 strictly excluded). Tests added: RouterPreferencesTest, ToolCallProbeAndUnifiedPoolTest. CI run 43719307999 on
  49. · Kildelinks · GitHub search · Z.ai · pulls · AlexsJones/llmfit · @jayhemnani9910
    Uddrag
    with Multi-head Latent Attention (DeepSeek-V2/V3, Kimi-K2, Moonlight, Kimi-VL, GLM-4.7-Flash) don't keep per-head K and V. They cache one latent of `kv_lora_rank + qk_rope_head_dim` = 512
  50. · Kildelinks · GitHub search · Z.ai · commits · Nitjsefnie/claudit · @github-actions[bot]
    Uddrag
    t, openai/gpt-5.2-codex, openai/gpt-oss-120b, openai/gpt-oss-20b, openai/gpt-oss-safeguard-20b, tencent/hy3-preview, xiaomi/mimo-v2.5, xiaomi/mimo-v2.5-pro, z-ai/glm-4.7-flash, z-ai/glm-5.3-prime
  51. · Kildelinks · GitHub search · Z.ai · commits · Nitjsefnie/ai-researcher · @github-actions[bot]
    Uddrag
    GLM-4.7-Flash (Reasoning) [Z AI]
  52. · Kildelinks · GitHub search · Z.ai · pulls · vllm-project/vllm · @hyjuunn
    Uddrag
    export MODEL=zai-org/GLM-4.7-Flash
  53. · Kildelinks · GitHub search · Z.ai · pulls · gorfreee/clasica_madrid · @gorfreee
    Uddrag
    | IA: requests por modelo | cloudflare:@cf/zai-org/glm-4.7-flash: 5, gemini:gemini-3-flash-preview: 2, gemini:gemini-3.1-flash-lite: 1, gemini:gemini-3.5-flash: 2, gemini:gemini-3.5-flash-lite: 2, gemini:gemini-3.6-flash: 2, gemini:gemini-3.7-flash: 2, gemini:gemini-3.8-flash: 2, gemini:gemma-4-26b-a4b-it: 1, gemini:gemma-4-31b-it:
  54. · Kildelinks · GitHub search · Z.ai · pulls · Prool-bot/ngo-grants-facebook · @Prool-bot
    Uddrag
    on explicit Workers-Free allowlist (also includes unselected free GLM-4.7-Flash; no automatic fallback or paid usage).
  55. · Kildelinks · GitHub search · Z.ai · pulls · HockerAGI/hocker.one · @HockerAGI
    Uddrag
    proof. Reuses existing `hocker-agi` gateway and fixed native `@cf/zai-org/glm-4.7-flash`; no Worker/Pages/DNS/SQL, no account spend configuration or live inference. Cloudflare account billing permissions returned 403 and
  56. · Kildelinks · GitHub search · Z.ai · commits · rodrigueztaylor87/rod · @rodrigueztaylor87
    Uddrag
    本文详解GLM-4.7-Flash在4张RTX 4090 D上的低成本、高性能部署实践，涵盖MoE架构优势、OpenAI兼容API集成、日志驱动故障诊断、配置热更新及典型并发与服务隔离问题排查。重点突出其针对中文场景的原生理解能力、低显存占用（单卡24GB可运行）、高吞吐推理（115 tokens/s）及开箱即用特性，适用于文案生成、知识问答、技术文档辅助等真实业务。
  57. · Kildelinks · GitHub search · Z.ai · commits · Trollz1004/ANTIGRAVITY · @Trollz1004
  58. · Kildelinks · GitHub search · Z.ai · commits · Nitjsefnie/ai-researcher · @github-actions[bot]
    Uddrag
    - GLM-4.7-Flash (Reasoning)
  59. · Kildelinks · GitHub search · Z.ai · commits · AmsyarAmran/Studyhub-stpm · @AmsyarAmran
    Uddrag
    Set Syar model to GLM-4.7-Flash Set Syar model to GLM-4.7-Flash
  60. · Kildelinks · GitHub search · Z.ai · commits · AmsyarAmran/Studyhub-stpm · @AmsyarAmran
    Uddrag
    Use GLM-4.7-Flash as Syar default model
  61. · Kildelinks · GitHub search · Z.ai · pulls · vllm-project/vllm · @lihzamd
    Uddrag
    | Model | GLM-4.7-Flash (`Glm4MoeLiteForCausalLM`) |
  62. · Kildelinks · GitHub search · Z.ai · pulls · HockerAGI/hocker.one · @HockerAGI
    Uddrag
    E2E en Preview; `agi_messages` de Supabase registró modelo `@cf/zai-org/glm-4.7-flash`, respuesta `LISTO`, mismo trace ID e inferencia real. No se renueva artificialmente esa prueba.
  63. · Kildelinks · GitHub search · Z.ai · pulls · HockerAGI/hocker.one · @HockerAGI
    Uddrag
    Supabase conservó la respuesta `LISTO`, route/provider `cloudflare-workers-ai`, modelo `@cf/zai-org/glm-4.7-flash`, `1481` tokens entrada/`3` salida y trace ID `9d09f050-695f-4628-9d7b-3fd35b2d97be`, un solo intento exitoso y ~740 ms de
  64. · Kildelinks · GitHub search · Z.ai · pulls · vllm-project/vllm · @Pb314314
  65. · Kildelinks · GitHub search · Z.ai · pulls · hale-lang/voice · @rileyr
    Uddrag
    put their thinking inline: the R1 distill, Qwen3-0.6B, GLM-4.7-Flash, MiMo's distill, OpenReasoning-Nemotron and Phi-4-mini-reasoning. vLLM v0.29.0 registers `deepseek_r1`, `qwen3`, `glm47` and `mimo`; each model needs
  66. · Kildelinks · GitHub search · Z.ai · pulls · eltmon/overdeck · @eltmon
    Uddrag
    here. On 1.5.2, `models.json` received only `glm-5.3-flashx`, `glm-5.1`, `glm-4.7-flash`, `mimo-v2.6-pro-ultraspeed` and `grok-build-0.1`.
  67. · Kildelinks · GitHub search · Z.ai · pulls · JoshCLWren/comic-pile · @JoshCLWren
    Uddrag
    Cloudflare Workers AI Free adapter and workers **119** (`@cf/zai-org/glm-4.7-flash`) and **120** (`@cf/google/gemma-4-26b-a4b-it`). These require `CLOUDFLARE_API_TOKEN` repository secret and `CLOUDFLARE_ACCOUNT_ID` repository variable. Absent credentials, workers remain
  68. · Kildelinks · GitHub search · Z.ai · pulls · anomalyco/models.dev · @models-dev-automation[bot]
    Uddrag
    - updated: `/home/runner/_work/models.dev/models.dev/providers/nano-gpt/models/z-ai/glm-4.7-flash:thinking.toml`
  69. · Kildelinks · GitHub search · Z.ai · pulls · Trollz1004/ANTIGRAVITY · @Trollz1004
  70. · Kildelinks · GitHub search · Z.ai · pulls · dannywise093-crypto/ATLAS · @dannywise093-crypto
    Uddrag
    - Connect the open-weight GLM-4.7-Flash model as ATLAS's first hosted foundation model.
  71. · Kildelinks · GitHub search · Z.ai · pulls · brokespace/seal · @brokespace
    Uddrag
    - **Structured output.** `glm-4.7-flash` is not on Workers AI's JSON Mode list, so the schema is enforced by the host: lenient JSON extraction, zod validation
  72. · Kildelinks · GitHub search · Z.ai · pulls · smg-project/smg · @hello-alexmcc
  73. · Kildelinks · GitHub search · Z.ai · pulls · laurofiorenzani/jarvis · @laurofiorenzani
    Uddrag
    The chat router now hard-allows only `@cf/google/gemma-4-26b-a4b-it` and `@cf/zai-org/glm-4.7-flash` for the Cloudflare Workers Free pool. Other models, including paid-only Kimi/GLM5 and speech/deprecated models, are rejected
  74. · Kildelinks · GitHub search · Z.ai · pulls · ai-advokat/ai-advokat.github.io · @ai-advokat
    Uddrag
    the timed-out Qwen recovery model with the already-supported GLM-4.7-Flash fallback with thinking disabled
  75. · Kildelinks · GitHub search · Z.ai · pulls · ACROWN-System/AstroCrown-Web · @ASDBradford-HAAN
    Uddrag
    latest heartbeat. Groq `gpt-oss-120b`, Gemini `gemini-3.8-flash` and Cloudflare `glm-4.7-flash` remain model/contract-specific blocked. Mistral remains deferred after repeated 429s; Cerebras is blocked under the no-payment-method rule;
  76. · Kildelinks · GitHub search · Z.ai · pulls · szl-holdings/a11oy · @stephenlutar2-hash
    Uddrag
    Both `MODEL_CANDIDATES` (`@cf/zai-org/glm-4.7-flash`, `@cf/google/gemma-4-26b-a4b-it`) are reasoning models that emit hidden `reasoning_content` before `message.content`. With the probe's `max_new_tokens: 96` (sent as `max_completion_tokens`), the entire completion budget
  77. · Kildelinks · GitHub search · Z.ai · pulls · spiceai/cookbook · @spiceemma
    Uddrag
    Two corrections specific to GLM: GLM-4.7-Flash's GGUF (the recipe's default file, and the abliterated variant's) declares `general.architecture = deepseek2`, not `glm4moe`. So the intro and
  78. · Kildelinks · GitHub search · Z.ai · pulls · methaneex-oss/AURIONS · @methaneex-oss
    Uddrag
    - Cloudflare GLM-4.7-Flash and Gemma 4 26B model pages: function calling; `max_tokens` deprecated in favor of `max_completion_tokens`; GLM documents `tool_choice: required`.
  79. · Kildelinks · GitHub search · Z.ai · pulls · brokespace/seal · @brokespace
    Uddrag
    Cloudflare Workers AI with the account's saved models (`@cf/zai-org/glm-4.7-flash`, fallback `@cf/google/gemma-4-26b-a4b-it`). Nothing is rewritten on disk, so current users keep working and a downgrade still
  80. · Kildelinks · GitHub search · Z.ai · pulls · SonicStar-Solutions/agentic-inbox · @cvbutani
    Uddrag
    - **Injection scanner (#30).** Default is now `@cf/zai-org/glm-4.7-flash` (ID confirmed against the Workers AI catalog). A reply with no usable verdict now **fails closed** (it
  81. · Kildelinks · GitHub search · Z.ai · pulls · diazMelgarejo/Perpetua-Tools · @diazMelgarejo
  82. · Kildelinks · GitHub search · Z.ai · commits · aphrody-labs/aphrody · @aphrody-dev
    Uddrag
    packages/design tokens; verified live: pong served by workers-ai glm-4.7-flash after a paid-only model was skipped
  83. · Kildelinks · GitHub search · Z.ai · pulls · sgl-project/sglang · @bvolpato
    Uddrag
    "stop"`. This happened with Llama-3.2-1B using `pythonic` and GLM-4.7-Flash using `glm47`. Streaming requests already returned the calls in `message.tool_calls`.
  84. · Kildelinks · GitHub search · Z.ai · pulls · eaitbrahim/NarratoFlow · @eaitbrahim
    Uddrag
    - llm row: local Ollama GLM-4.7-Flash Q4_K_M, $0. catt row: linux/arm64 Docker container on the same Mac, CPU. No pyproject extra (the universal uv lock
  85. · Kildelinks · GitHub search · Z.ai · pulls · diazMelgarejo/orama-system · @diazMelgarejo
  86. · Kildelinks · GitHub search · Z.ai · pulls · chadkhalil21-svg/Plenum-Scientific-Intelligence · @chadkhalil21-svg
    Uddrag
    adds the upstream native `[ai]` binding to free-eligible `@cf/zai-org/glm-4.7-flash`, and routes ORIGINAL owner-authorized OpenHands issue jobs through a narrowly compatible endpoint. No new deployment, host,
  87. · Kildelinks · GitHub search · Z.ai · pulls · JoshCLWren/comic-pile · @JoshCLWren
    Uddrag
    Model: glm-4.7-flash Source: z-ai
  88. · Kildelinks · GitHub search · Z.ai · commits · lettroi/safeshell-pro · @lettroi
    Uddrag
    - 5 enabled free models (glm-4.7-flash disabled by default due to upstream 429)
  89. · Kildelinks · GitHub search · Z.ai · commits · 1essalion1-sudo/Nexora-exotic · @1essalion1-sudo
    Uddrag
    Document Neora website and GLM-4.7-Flash model credit
  90. · Kildelinks · GitHub search · Z.ai · pulls · salamou1944/Salamou-31 · @salamou1944
    Uddrag
    probe's default `@cf/meta/llama-3.1-8b-instruct` model, deprecated on 2026-05-30, with `@cf/zai-org/glm-4.7-flash`.
  91. · Kildelinks · GitHub search · Z.ai · pulls · luongnv89/freetokens · @luongnv89
  92. · Kildelinks · GitHub search · Z.ai · pulls · diazMelgarejo/orama-system · @diazMelgarejo
    Uddrag
    The runner still starts at `glm-4.7-flash`. Success, billing code 1113, credentials, throttle, and uncertain execution stop the chain. A verified model miss is what advances
  93. · Kildelinks · GitHub search · Z.ai · pulls · chadkhalil21-svg/Plenum-Scientific-Intelligence · @chadkhalil21-svg
    Uddrag
    git blob), same signed original Cloudflare Worker, same `glm-4.7-flash` model and original Durable Object 4,000/day local usage bound. Include actual original public source, original bounded
  94. · Kildelinks · GitHub search · Z.ai · pulls · LevenST24/galgame2voice · @LevenST24
  95. · Kildelinks · GitHub search · Z.ai · pulls · thisishaidee/kivora · @thisishaidee
    Uddrag
    adapter for Cloudflare Workers AI using the free-tier `@cf/zai-org/glm-4.7-flash` model.
  96. · Kildelinks · GitHub search · Z.ai · pulls · luongnv89/freetokens · @luongnv89
    Uddrag
    | bumped | — | — | \| GLM-4.7-Flash \| Free \| Free \| Free \| Free \| |
  97. · Kildelinks · GitHub search · Z.ai · pulls · HYGON-AI/vllm-plugin-das · @alexanderbin123
    Uddrag
    Additional GLM-4.7-Flash BF16 TP2 validation on gfx936, same frozen requests, official TRITON_MLA/default Graph/prefix: official baseline and candidate both score **0/32 raw, 30/32 format-normalized**, but the
  98. · Kildelinks · GitHub search · Z.ai · pulls · GoDanceAi/godance · @SummerTiger
    Uddrag
    id not in the provider's model list. ernie-5.1, glm-4.7-flash: text only. kimi-k2.6, minimax-m2.7: vision documented for a newer family only. qwen-3.6, qwen-3.6-coder: the documented ids
  99. · Kildelinks · GitHub search · Z.ai · pulls · sgl-project/sglang · @ami-GS
    Uddrag
    Found while serving `zai-org/GLM-4.7-Flash` with default arguments on an `sm100` target, where the first prefill request immediately crashes in FlashInfer:
  100. · Kildelinks · GitHub search · Z.ai · pulls · quiquandon-oss/PulseWorkerV2 · @quiquandon-oss
    Uddrag
    the 2026-11-14 scope lock: RSS/Atom/official feeds only, optional GLM-4.7-flash with a hard daily Neuron budget, and no new providers, models or paid dependencies. The design
  101. · Kildelinks · GitHub search · Z.ai · pulls · can1357/oh-my-pi · @andrebrait
    Uddrag
    refusal without consent, immediate running state, a real GLM-4.7-Flash cited result, independently confirmed application, idempotent replay, and getter/state equality. Temporary credentials and fixtures were removed.
  102. · Kildelinks · GitHub search · Z.ai · pulls · HockerAGI/hocker.one · @HockerAGI
    Uddrag
    el adaptador **ya existente** `cloudflareWorkersAiProvider.complete` con modelo fijo `@cf/zai-org/glm-4.7-flash`, 384 output tokens, timeout, 1 intento, sin tools, sin web search, sin Vercel/OpenAI fallback. Resultado/uso/hash se
  103. · Kildelinks · GitHub search · Z.ai · pulls · vllm-project/vllm · @umarkovi-amd
    Uddrag
    inactive since #46009. On Radeon Pro W7900 with GLM-4.7-Flash at concurrency 1, this restores 4.5–10.8 % output throughput and 10–13 % TTFT. Related: #46009, #46381
  104. · Kildelinks · GitHub search · Z.ai · pulls · anomalyco/models.dev · @models-dev-automation[bot]
  105. · Kildelinks · GitHub search · Z.ai · pulls · anomalyco/models.dev · @KevRojo
  106. · Kildelinks · GitHub search · Z.ai · pulls · afjcjsbx/picoclaw · @trufae
    Uddrag
    model invocation. The local configuration names `nemo` → `glm-4.7-flash`, but that model was not used for an end-to-end image-description test. The model on the gateway
Specifikationer

GLM-4.7 Flash er en model fra Z.ai med åbne vægte, først registreret den 2026-09-17; lanceringsdatoen er ikke fastlagt. Kontekstvinduet er 202.752 tokens.

Kontekstvindue
202.752
Vægte
Åbne

Spørgsmål

Hvor stort er kontekstvinduet for GLM-4.7 Flash?

Kontekstvinduet er 202.752 tokens.

Er vægtene for GLM-4.7 Flash offentligt tilgængelige?

Ja. Z.ai offentliggør vægtene.

Hvornår blev GLM-4.7 Flash lanceret?

Lanceringsdatoen er ikke fastlagt. Signal Forge registrerede modellen første gang den 2026-09-17.