GLM-5.3-Flash
The small, fast sibling of GLM-5.3, released with open weights under the MIT license and priced very low on Z.ai's API. It clears 40 on the Artificial Analysis Intelligence Index at a cost of $0.28 per task.
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 #3, Speed #14 (tied), Reasoning #18.
Strengths and weaknesses
- Very low price ($0.15 / $0.50 per million tokens) and the third-best value in this list
- 65.8% on ARC-AGI-2 at max effort and 35.8% on Terminal-Bench 4.0 in Claude Code
- About 6.6 million downloads on Hugging Face
- Behind GLM-5.3 on most benchmarks (Intelligence Index 41.8 vs 44.8)
- In the lower half of this list on the LMArena text and writing leaderboards
Best for
- A cheap coding helper
- Self-hosting
- High-volume tasks
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)1,473.8Arena score 1473.8 (95% CI 1469.0 to 1478.7); 32,925 votes; entry "glm-5.3-flash"; position 41 of 416 on the full board
- LMArena creative writing1,433.7Arena score 1433.7 (95% CI 1425.5 to 1441.8); 6,983 votes; entry "glm-5.3-flash"; position 68 of 414 on the full board
- LMArena WebDev (coding)1,605.5Arena score 1605.5 (95% CI 1598.8 to 1612.1); 17,579 votes; entry "GLM-5.3-Flash"; position 27 of 144 on the full board
- LMArena vision1,273.3Arena score 1273.3 (95% CI 1266.3 to 1280.4); 11,157 votes; entry "glm-5.3-flash"; position 35 of 165 on the full board
- LMArena Agent Arena-0.012IPS score -0.0116 (95% CI -0.0185 to -0.0048); 87,298 sessions; entry "GLM-5.3-Flash"; position 33 of 52 on the full board
- Artificial Analysis Intelligence Index41.841.8 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 42), setting "GLM-5.3-Flash"
- Humanity's Last Exam (Artificial Analysis)39.9%39.9% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "GLM-5.3-Flash"
- GPQA Diamond, graduate-level science91.2%91.2% on GPQA Diamond (AA; no longer part of the AA index since v4.2), setting "GLM-5.3-Flash"
- Terminal-Bench 4.0 (Artificial Analysis run)32.8%32.8% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "GLM-5.3-Flash"
- Long-context reasoning (AA-LCR)80.0%80.0% on AA-LCR (long-context reasoning), setting "GLM-5.3-Flash"
- AA-Omniscience, accuracy minus hallucination7.57.5 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "GLM-5.3-Flash"
- Output speed, tokens per second71 t/s71 output tokens per second (AA median), row "GLM-5.3-Flash"
- Cost per Artificial Analysis test task$0.28$0.28 to run one AA Intelligence Index task, row "GLM-5.3-Flash"
- Terminal-Bench 4.0 leaderboard35.8%35.8% resolved on Terminal-Bench 4.0 (plus or minus 3.5, 95% CI) with Claude Code at none effort
- ARC-AGI-2 puzzles65.8%65.83% on ARC-AGI-2 (semi-private set), best listed setting "GLM-5.3-Flash (Max)", $0.09 per task
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.