GLM-5.3 Flash Benchmarks Hit GPT-5.6 Territory at Fractional Cost
Around Aug 27, a wave of third-party tests of GLM-5.3 Flash (nicknamed the "bull model") converged on one conclusion: frontier-level capability at extremely low cost, sitting on the Pareto frontier of price-performance—and reportedly able to recreate a 3D game in an hour.
Confirmed
- Local deployment: @fundingsecured ran it on a single DGX Station GB300 at 206 tokens/s single-stream with 1M-token context (m1). Figures relayed by Alex Ellis show 881 tok/s at C=64 and 232 tok/s at C=1 on 2x DGX Station (TP=2) without tuning (m2/m4 are duplicate relays).
- Sentdex tested an upcoming release: 180 tok/s at native precision, excellent concurrency scaling, and vision support; he said he won't return to closed APIs unless full 5.3 is stronger (m3).
- Benchmarks and cost: Zain Hasan reported a score of 58 on the AA Agentic Index—on par with GPT-5.6 Sol—at $0.09 per task, Luna-tier cheap (m5). Per ArtificialAnalysis, it scores 57 intelligence at $0.045 per task, on the Pareto frontier and beating rivals including DeepSeek V4 Flash (m7).
- Hands-on: @AIandDesign called it "really good" after extended LM Studio testing, costing only 20 cents so far (m6); @卡尔的AI沃茨 reported recreating a 3D game in an hour (m7).
Unconfirmed
- All benchmark and cost figures come from individual users and third-party indices; no official numbers appear in the posts, and Sentdex's build is a pre-release version whose final behavior remains to be verified.
Why it matters
- If these tests hold, GLM-5.3 Flash would deliver near-GPT-5.6 intelligence at cent-level per-task cost, locally deployable with long context and vision—potentially disrupting API pricing and the low-cost open-model landscape.
2026-08-27 ~ 2026-08-28 · 7 related posts
- Episode 1: Zhipu GLM-5.3 Spotted in Codebase, Post-Training Targets K3 and GPT-5.6(2026-08-03, 5 posts)
- Episode 2: Zhipu AI Releases GLM-5.3: Same Base Model, Post-Training Drives a Big Leap(2026-08-14, 51 posts)
- Episode 3: AI Weekly: Grok, Gemini Updates, and Anthropic's Secret Model(2026-08-14, 2 posts)
- Episode 4: Alibaba, Zhipu and DeepSeek Unveil New Models on the Same Day(2026-08-15, 3 posts)
- Episode 5: Zhipu Launches GLM-5.3 API with 50% Coding Boost(2026-08-19, 4 posts)
- Episode 6: GLM-5.3 Scores 60 on AA Intelligence Index, Tops Agentic Index(2026-08-19, 12 posts)
- Episode 7: Zhipu GLM-5.3 Tops DeepSWE at a Fraction of the Cost(2026-08-20, 4 posts)
- Episode 8: Zhipu Open-Sources GLM-5.3-Flash: 320B MoE Multimodal at 1/10th the Cost(2026-08-25, 48 posts)
- Episode 9: GLM-5.3 Flash Benchmarks Hit GPT-5.6 Territory at Fractional Cost(2026-08-27, 7 posts)
- Episode 10: GLM-5.3-Flash Ranks Third Among Open Models at Half of Qwen's Cost(2026-08-27, 2 posts)
- Episode 11: GLM 5.3 Flash Gets 90% Discount Through September(2026-08-27, 2 posts)
- Episode 12: Zai Releases GLM-5.3 Weights with Tiered Access Citing Cybersecurity Capabilities(2026-08-27, 11 posts)
- Episode 13: GLM-5.3 Flash Benchmarks: Near-Frontier Quality at a Fraction of the Cost(2026-08-27, 12 posts)
- Episode 14: Z.ai Releases Open-Weight GLM-5.3 with Day-One Adoption Across Platforms(2026-08-29, 19 posts)
- Episode 15: Zhipu's GLM-5.3 Launches Exclusively on Baseten with Big Coding Gains(2026-09-05, 2 posts)
Primary sources
- GLM-5.3 Flash Review: GPT-5.6 Level Performance at Ultra-Low Cost — zainhas · 2026-08-27
- GLM 5.3 Flash Benchmark: Hits 881 tok/s on Dual DGX — teortaxesTex · 2026-08-27
- User tests GLM-5.3 Flash: excellent performance and extremely low cost — AIandDesign · 2026-08-27
- [source] GLM-5.3-Flash Review: Beats DeepSeek V4, Clones 3D Game in 1 Hour — 卡尔的AI沃茨 · 2026-08-27
- [source] Local Deployment of GLM-5.3-Flash: 206 tok/s and 1M Context on DGX Station — funding__secured · 2026-08-27
- [source] GLM 5.3 Flash Impresses with 180 tok/s Speed and Vision Support — Sentdex · 2026-08-28
1 near-duplicate retellings: HankYeomans