GLM 4.5 Air
z-ai/glm-4.5-air- Z.ai
- Family · GLM 4 Air
- Released 28 Jul 2025
- Open weights
- Text → Text
- Tools
Not on OSMEXO yet
Spec & list price
Context128K
Max output96K
Input / 1M$0.13
Output / 1M$0.856.5× input
Scores · standing
Intelligence11.1#232/490
Usage
Tokens / day798.4M▼ 15% 7d
Usage rank#204
Requests / day199.5K
Tool calls / req0.37
Per request · 30 days
Avg input7.0K0.5× median
Avg output2710.4× median
Cache hit77%about median
Reasoning share49%0.9× median
Tool-call failures2.5%1.3× median
Consumption
Peak 5.5B 14 OctLow 355.7M 18 OctRank today #204
News & trends
News · latest from Z.ai
- Hermes and REA and GLM 5.3 = Reverse engineering TikTok app for less than $20scraping.club · 29 Oct
- GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behaviorlesswrong.com · 27 Oct
- Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)blog.cloudflare.com · 26 Oct
- ReplaySSM: Rethinking State Storage for Faster Hybrid-Model DecodingNVIDIA · 12 Oct
Benchmarks
bar = share of the top score · Benchmark data: Artificial AnalysisSpeed · median across 3 providers, latest day
Output speed#227/33448 tok/s1.3× slower than median
Fastest provider68 tok/sbest of 3 providers
First token#1721.13 sabout median
#13 cheapest of 33 with a similar score
Value
Intelligence Index
Blended price per 1M tokens (log)
Technical details
- Tokenizer
- Other
- Knowledge cutoff
- 2024-12-31
- License
- MIT
- Parameters
- 110B
- Moderation
- none
- Downloads · likes
- 1.3M · 649
Request parameters 13
- tools
- tool_choice
- reasoning
- include_reasoning
- temperature
- top_p
- top_k
- max_tokens
- stop
- seed
+3 more
- frequency_penalty
- presence_penalty
- repetition_penalty
Interesting facts
- Weekdays average 2.0B tokens a day against 651.1M on weekends.
- 1.5× the output room of similar models (98.3K against a 65.5K median).
When to use
Best for
- Low-cost math and science reasoning
- Agents that call tools
- Self-hosted open-weight deployments
- Code generation on a budget
Think twice for
- Image understanding
- Strict structured-output pipelines
- Hardest terminal-style agent runs
- Frontier-level general reasoning
Research papers
- Reconstruction-Optimized Expert Pruning for Mixture-of-Experts Language Models19 Oct
- Epistemic Policy Divergence in Multi-Turn LLM Contamination: A Protocol-Gradient Investigation28 Sep
- What LLM Forecasters Know but Don't Say: Probing Internal Representations for Calibration and Faithfulness9 Jul
- CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents6 Jul
Alternatives
other vendors · similar capability- 12.9
- 8.5
- 11.6
- 13.1
Want GLM 4.5 Air on OSMEXO?
Sign in to request it. Requests are reviewed; a request does not guarantee availability.