Your sovereign, private GLM 5.3 Flash
GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series: 320B total with just 18B active parameters, outperforming GLM-5.2 across benchmarks at one-tenth the price while approaching Claude Opus 4.8 on coding and agentic tasks. A redesigned hybrid sparse+linear attention architecture sharply cuts long-context serving costs, with a 1M-token context and configurable reasoning effort. A reasoning model in the medium tier, served on ToothFairyAI's AU, EU and US endpoints — private by default, billed in US dollars per million tokens at the published rates below. No training on your data, no vendor lock-in.
How GLM 5.3 Flash compares
Independent intelligence index for GLM 5.3 Flash against every peer in the Medium tier (and the tier above where needed) — same source, same tests.
Latency — end-to-end seconds for a 500-token answer
Capability indexes
Domain-weighted agentic performance for GLM 5.3 Flash.
Capability index vs medium peers
The evaluations behind the index
Independent benchmark results for GLM 5.3 Flash — Elo scores as published; pass rates shown as percentages.
Per-evaluation economics
| Evaluation | Score | Time / task | Output tokens / task |
|---|---|---|---|
| AA-Briefcase | 1459.41 | 38.4 min | 131965 |
| GDPval-AA | 1640.78 | 36.7 min | 125884 |
| AutomationBench-AA | 60.4% | 5.9 min | 20219 |
| Terminal-Bench 4.0 | 32.8% | 52.7 min | 181054 |
| SciCode | 51.6% | 7.9 min | 27041 |
| Humanity's Last Exam | 39.9% | 12.6 min | 43214 |
| GDP.pdf | 15.4% | 4.6 min | 15701 |
| CritPt | 15.4% | 23.8 min | 81785 |
| AA-Omniscience | 7.47 | 0.3 min | 1159 |
| AA-LCR (long context) | 80.0% | 1.3 min | 4515 |
Run GLM 5.3 Flash privately
One key, one endpoint, three jurisdictions.
- Size tierMedium
- Serving routeStandard
- Input rate$0.42/1M
- Output rate$1.51/1M
- Cached input rate—
- Context window1049K tokens
- Max output16K tokens
- ReasoningYes — emits thinking traces
- Vision & video inputSupports image input
- Tool callingSupported
- Data residencyAU, EU and US endpoints — processing stays inside the region you call