GLM-5.3
Z.ai's open-weight flagship (MIT license) focused on complex software engineering, with a 1M-token context window and text-only input. It is one of the strongest open models for agentic coding in this list.
Z.ai (Zhipu)Language modelReleased August 2026Data checked October 18, 2026
This model is not in the current people's rankings: it is either new and still collecting votes, or ranked below the top 12. See the AI leaderboard for the models people rate highest.
On paper
Fixed-test rankings from the benchmarks tab: Value #9, Speed #12, Reasoning #16.
Strengths and weaknesses
- Highest Terminal-Bench 4.0 score of any open model on the tbench.ai board (41.8% in Claude Code)
- Artificial Analysis Intelligence Index of 44.8 under a permissive MIT license
- Moderate price ($1.40 / $4.40 per million tokens)
- Text-only input, so it cannot read images
- Mid-table on the LMArena text and creative writing leaderboards
Best for
- Open-weight coding agents
- Low-cost coding help
- Self-hosting for software work
Every score we have
Each bar compares this model with the best result any model on this page has on the same test.
- LMArena text arena (human votes)1,478.5Arena score 1478.5 (95% CI 1473.3 to 1483.8); 22,543 votes; entry "glm-5.3-max"; position 28 of 416 on the full board
- LMArena creative writing1,453.2Arena score 1453.2 (95% CI 1443.9 to 1462.5); 5,066 votes; entry "glm-5.3-max"; position 35 of 414 on the full board
- LMArena WebDev (coding)1,621.1Arena score 1621.1 (95% CI 1613.5 to 1628.8); 9,542 votes; entry "GLM-5.3 (Max)"; position 23 of 144 on the full board
- LMArena Agent Arena0.018IPS score 0.0178 (95% CI 0.0094 to 0.0262); 86,541 sessions; entry "GLM-5.3 (Max)"; position 24 of 52 on the full board
- Artificial Analysis Intelligence Index44.844.8 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 45), setting "GLM-5.3 (max)"
- Humanity's Last Exam (Artificial Analysis)42.3%42.3% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "GLM-5.3 (max)"
- GPQA Diamond, graduate-level science91.7%91.7% on GPQA Diamond (AA; no longer part of the AA index since v4.2), setting "GLM-5.3 (max)"
- Terminal-Bench 4.0 (Artificial Analysis run)41.9%41.9% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "GLM-5.3 (max)"
- Long-context reasoning (AA-LCR)79.7%79.7% on AA-LCR (long-context reasoning), setting "GLM-5.3 (max)"
- AA-Omniscience, accuracy minus hallucination14.314.3 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "GLM-5.3 (max)"
- Output speed, tokens per second81 t/s81 output tokens per second (AA median), row "GLM-5.3 (max)"
- Cost per Artificial Analysis test task$2.01$2.01 to run one AA Intelligence Index task, row "GLM-5.3 (max)"
- Artificial Analysis Coding Agent Index53.653.6 on the AA Coding Agent Index v1.5 (DeepSWE v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA), best listed setup "Opencode - GLM-5.3"
- Terminal-Bench 4.0 leaderboard41.8%41.8% resolved on Terminal-Bench 4.0 (plus or minus 3.2, 95% CI) with Claude Code at max effort
Compare with
- Claude Opus 5.5Anthropic's flagship for long-running agentic coding and knowledge work, and the model Anthropic suggests starting with for most jobs.
- Claude Sonnet 5.5Anthropic's mid-priced model, which Anthropic describes as its best mix of speed and intelligence.
- Claude Fable 5.1Anthropic's top tier for the hardest reasoning and long-horizon agent work, which Anthropic recommends when Opus 5.5 still falls short.
- Claude Haiku 5.5Anthropic's smallest and fastest current model, built for high-volume, latency-sensitive jobs like classification, extraction and routing.