# We shrank GLM-5.3-Flash by 68%, and it solved 86% of SWE-bench Verified

Tileward research report R-012. GLM-5.3-Flash is Z.ai's 321-billion-parameter model. In 16 bits its weights are 642.65 GB. Tileward's TWXH8 build is 204.45 GB, 68% smaller (3.14×), and runs on two B300 cards. Working as a coding agent it solved 431 of the 500 SWE-bench Verified tasks, and 429 when the whole run was repeated. The original was not run on SWE-bench, so the comparison with it is next-word agreement: TWXH8 picks the original's next word 88.8% of the time.

Page: https://tileward.com/research/a-smaller-coding-agent/

## How it was measured

The SWE-bench run was repeated in full with the same settings, to show how much a score moves between identical runs.

- **Builds:** [GLM-5.3-Flash in 16 bits](https://huggingface.co/zai-org/GLM-5.3-Flash-BF16) and two Tileward builds: TWX48 and TWXH8. TWXH8 is the build run on SWE-bench.
- **Machine:** two B300 cards, with stock vLLM serving TWXH8 from its own weights, every weight on the cards.
- **Size:** the weights files on disk.
- **SWE-bench Verified:** [SWE-bench Verified](https://openai.com/index/introducing-swe-bench-verified/), all 500 tasks, with [mini-SWE-agent](https://github.com/SWE-agent/mini-swe-agent): up to 250 steps, temperature 1.0, top-p 0.95, high reasoning effort, replies up to 16,384 tokens and a 65,536-token window. Every patch was graded by the official SWE-bench grader on the original task set. Tasks whose container failed to start were run again before grading; no task was run again to improve a score.
- **Agreement:** each build reads the same 32 passages of [WikiText-2](https://arxiv.org/abs/1609.07843) (16,352 token positions), and we count where its top next token is the original's. Every build was scored in one run against the same 16-bit original.

## Results

| Build | Weights | Against the original |
|---|---|---|
| GLM-5.3-Flash in 16 bits | 642.65 GB | The reference |
| Tileward TWX48 | 205.09 GB | 68% smaller (3.13×) |
| Tileward TWXH8 | 204.45 GB | 68% smaller (3.14×) |

SWE-bench Verified, TWXH8:

| Run | Solved | Patch submitted, not solved | No patch |
|---|---|---|---|
| First run | 431 of 500 (86.2%) | 61 | 8 |
| Repeat, same settings | 429 of 500 (85.8%) | 66 | 5 |

411 tasks were solved in both runs and 38 (7.6%) in only one, so a difference of a few tasks between runs of this build is noise. Most tasks with no patch ran past the 65,536-token window.

Next-word agreement with the original:

| Build | Next-word agreement | KL distance from the original |
|---|---|---|
| Tileward TWX48 | 87.3% | 0.115 |
| Tileward TWXH8 | 88.8% | 0.088 |

On how well each predicts the text, the two builds are level with each other: the difference is within chance.

## Related

- [GLM-5.3-Flash eval card](https://tileward.com/evals/#compressed-model-glm): the result in one screen
- [GLM-5.3-Flash on Hugging Face](https://huggingface.co/zai-org/GLM-5.3-Flash): the published model
- [Talk to an engineer](https://calendar.app.google/hg2DzFTxKJgwFWm19): bring the code your agents work on and the cards you run on
- [Research](https://tileward.com/research/): every report