GLM-5.1. is Now Available on Ajelix AI Chat
Dicover other articles
Z.ai’s next-generation agentic flagship joins the Ajelix model lineup with significantly stronger coding capability, long-horizon autonomy, and a #1 open-source ranking on SWE-Bench Pro.
Starting today, Ajelix users can select GLM-5.1 as their model of choice inside Ajelix Chat. Released on April 7, 2026, by Z.ai. GLM-5.1 is a post-training refinement of the acclaimed GLM-5 architecture, purpose-built to push the ceiling on long-horizon agentic engineering. It is the first open-source model to claim the top spot on SWE-Bench Pro, surpassing GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro on one of the most demanding professional coding benchmarks in existence.
“GLM-5.1 is simply the better choice for long agentic workflows. After our internal tests, the results speak for themselves: it’s more accurate, it performs edits more precisely, and it brings a noticeably bigger technical capability across the board. That’s exactly why we added it to our model selection: so our customers can tackle more advanced, demanding workflows with confidence.” – Arturs, CTO at Ajelix
What Is GLM-5.1?
GLM-5.1 shares the same 744-billion-parameter Mixture-of-Experts architecture as GLM-5, with approximately 40 billion active parameters per token and a 200,000-token context window. What sets it apart is a fundamentally different approach to how it tackles hard problems over time.
Previous models, including GLM-5, tend to front-load their effort: they apply familiar strategies, make early gains, and then plateau. GLM-5.1 is designed to stay effective over much longer horizons. It breaks down complex problems, runs experiments, reads results, identifies blockers with precision, and revises its strategy through repeated iterations. The model can sustain fully autonomous execution for up to 8 hours on a single task, completing a full “experiment → analyze → optimize” loop without human intervention.
Benchmark Performance
GLM-5.1 ranks #1 in open source and #3 globally across SWE-Bench Pro, Terminal-Bench 2.0, and NL2Repo. Selected results against leading models:
| Benchmark | GLM-5.1 | GLM-5 | DeepSeek-V3.2 | Kimi K2.5 | Claude Opus 4.6 |
|---|---|---|---|---|---|
| SWE-Bench Pro | 58.4 | 55.1 | — | 53.8 | 57.3 |
| HLE (w/ Tools) | 52.3 | 50.4 | 40.8 | 51.8 | 53.1 |
| Terminal-Bench 2.0 | 63.5 | 56.2 | 39.3 | 50.8 | 65.4 |
| NL2Repo | 42.7 | 35.9 | — | 32.0 | 49.8 |
| BrowseComp (w/ Context) | 79.3 | 75.9 | 67.6 | 74.9 | 84.0 |
| GPQA-Diamond | 86.2 | 86.0 | 82.4 | 87.6 | 91.3 |
| CyberGym | 68.7 | 48.3 | 17.3 | 41.3 | 66.6 |
The Terminal-Bench 2.0 and NL2Repo numbers are especially relevant for Ajelix workflows. Terminal-Bench measures autonomous real-world terminal task completion; NL2Repo tests full repository generation from natural language. GLM-5.1 shows a 13% improvement over GLM-5 on Terminal-Bench 2.0 and a 19% leap on NL2Repo, both directly translating to better performance on the agentic, code-heavy features Ajelix users rely on every day.
Why We Added GLM-5.1 to Ajelix?
Our model selection criteria have always centered on three things: capability breadth, reasoning reliability, and open availability. GLM-5.1 clears all three bars decisively.
The architectural foundation is identical to GLM-5, a Mixture-of-Experts model with DeepSeek Sparse Attention (DSA) for efficient long-context handling up to 200K tokens. What GLM-5.1 adds on top is a rigorously improved post-training process: multi-task supervised fine-tuning and reinforcement learning specifically tuned for long-horizon execution. This is where it earns its place alongside the other frontier models in our lineup.
“Providing our users with the newest and most capable technologies is one of the core foundations of our mission at Ajelix. Adding GLM-5.1 is a direct expression of that commitment, and you can expect more updates from us as the model landscape continues to evolve.” – Agnese, COO at Ajelix
When to Choose GLM-5.1 in Ajelix
Not every task needs the most powerful model available. But when your work hits any of the following scenarios, GLM-5.1 is the right choice.
- Long agentic workflows & multi-step automation. GLM-5.1’s core advantage is its ability to remain productive over extended sessions: revising strategy, re-running experiments, and sustaining progress across hundreds of iterations. For complex Ajelix workflows involving multiple tools, branching decisions, and iterative outputs, this translates to better end results with fewer interruptions.
- Complex code generation & technical tasks. A 58.4 on SWE-Bench Pro #1 globally among open-source models speaks directly to coding quality. For users working with data pipelines, API integrations, script generation, formula logic, or system configurations, GLM-5.1 delivers more accurate first-pass results and stronger error diagnosis than its predecessor.
- Full repository & project generation. NL2Repo evaluates a model’s ability to generate complete, functional repositories from natural language descriptions. GLM-5.1’s 19% improvement over GLM-5 here means better, more structured code output when building out entire project scaffolds or application skeletons from Ajelix.
- Deep document analysis & web research. With a BrowseComp (with Context Management) score of 79.3, GLM-5.1 maintains coherent reasoning across dense multi-document analyses and extended web research sessions. It holds context reliably when drawing conclusions that depend on information spread across a long conversation or multiple file sources.
- Landing pages, visual assets & front-end code. GLM-5.1’s stronger code generation carries directly into front-end output quality. Complex CSS interactions, SVG graphics, multi-section layouts, and component-level detail come out cleaner, requiring fewer corrections before they’re production-ready.
How to Access GLM-5.1 in Ajelix
To select GLM-5.1, simply open the model dropdown in Ajelix Chat and choose it from the list. The model is available now for all plan users.
Agentic AI chat that helps you complete projects
