Compare / Kimi K3 vs GLM-5.3
Kimi K3 vs GLM-5.3
Which is cheaper, which is more capable, and which does better at each kind of work, from 39 results on 7 sources that measured both models.
Of the 39 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.
Results as of 16 November 2026 (catalogue release 2026-11-16-49bdfa045dee). See also Moonshot AI vs Z.ai, every model from both labs.
Kimi K3 or GLM-5.3? The short answer
Pick Kimi K3 if quality matters most: it scores higher (Spring Prompt overall 47 against 35). Pick GLM-5.3 if cost matters more: it costs 3.5 times less per game of bullet chess on BulletBench. GLM-5.3 is also faster.
- Spring Prompt overall: Kimi K3 47, GLM-5.3 35 (out of 100; how the score works).
- GLM-5.3 is cheaper to run: $0.016 against $0.054 per game of bullet chess on BulletBench (3.5 times less).
- GLM-5.3 scores higher on the Artificial Analysis Intelligence Index: 44.8 against 43.6.
- GLM-5.3 writes faster: 87 tokens a second against 41.
- GLM-5.3 is newer: released 14 Aug 2026, against 16 Jul 2026 for Kimi K3.
- On our own benchmarks, on the results both models have: Kimi K3 does better for marketing planning.
- On third-party benchmarks, on the results both models have: Kimi K3 does better for reasoning and knowledge and sticking to the facts; GLM-5.3 for agents and tool use and professional work.
Head to head
Every fact side by side
| Kimi K3 | GLM-5.3 | |
|---|---|---|
| Price per million tokens | $3.00 in · $15.00 out | $1.40 in · $4.40 out |
| Blended (3 in : 1 out) | $6.00 | $2.15 |
| Cheapest host, blended | $2.47 (Relace, FP4) | $0.40 (Wafer, FP4) |
| Cost per game of bullet chess (BulletBench) | $0.054 | $0.016 |
| Cached input, per million | $0.30 cached input | $0.26 cached input |
| Intelligence Index | 43.6 | 44.8 |
| Spring Prompt overall | 47 | 35 |
| Output speed, tokens a second | 41 | 87 |
| Context window | 1,048,576 tokens | 1,048,576 tokens |
| Released | 16 Jul 2026 | 14 Aug 2026 (newer) |
| Weights | Open | Open |
| Developer based in | China | China |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.
From our research
Every shared result
Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.
| Measure | Kimi K3 | GLM-5.3 |
|---|---|---|
| BulletBench measured by Spring Prompt · 3 results | ||
| Cost per game · Bullet 60sUS dollars, lower is better 55 · 44 of 59 measured | $0.0540low reasoning | $0.0156 |
| Ladder Elo · Bullet 60sladder Elo, higher is better 56 · 56 of 59 measured | 33low reasoning | 33 |
| Games lost on time · Bullet 60s% of games, lower is better 58 · 53 of 59 measured | 91.7%low reasoning | 83.3%low reasoning |
| ROASBench measured by Spring Prompt · 3 results | ||
| Overall scorescore out of 100, higher is better 17 · 30 of 33 measured | 44.1 | 24.8 |
| Audience scorescore out of 100, higher is better 15 · 26 of 33 measured | 65.1 | 51.0 |
| Business scorescore out of 100, higher is better 17 · 30 of 33 measured | 40.2 | 12.9 |
| APEX-Agents reported by APEX-Agents · 1 result | ||
| Tasks passed% of tasks, higher is better 21 · 13 of 37 measured | 50.6% | 56.6% |
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results | ||
| Overall · TextArena rating, higher is better 13 · 24 of 387 measured | 1,488max reasoning | 1,478max reasoning |
| Overall · Text factualityArena rating, higher is better 20 · 29 of 163 measured | 1,472max reasoning | 1,467max reasoning |
| Business, management and finance · TextArena rating, higher is better 12 · 26 of 380 measured | 1,487max reasoning | 1,475max reasoning |
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better 19 · 17 of 249 measured | 43.6max reasoning | 44.8max reasoning |
| GPQA Diamond% of questions, higher is better 10 · 26 of 232 measured | 93.5%max reasoning | 91.7%max reasoning |
| Humanity's Last Exam% of questions, higher is better 20 · 38 of 250 measured | 46.9%max reasoning | 42.3%max reasoning |
| SimpleQA Verified (Epoch AI) reported by SimpleQA Verified (Epoch AI) · 1 result | ||
| Correct answers% of questions, higher is better 23 · 42 of 76 measured | 50.6%max reasoning | 41.0%max reasoning |
| Vending-Bench 2 reported by Vending-Bench 2 · 1 result | ||
| Money after a yearUS dollars, higher is better 24 · 8 of 58 measured | $5,165 | $8,164 |
Show all 39 results (24 more)
| Measure | Kimi K3 | GLM-5.3 |
|---|---|---|
| BulletBench measured by Spring Prompt · 2 results | ||
| Invalid moves · Bullet 60s% of moves, lower is better 1 · 1 of 59 measured | 0.0%low reasoning | 0.0% |
| Median move time · Bullet 60smilliseconds, lower is better 49 · 42 of 59 measured | 2.6 slow reasoning | 1.6 slow reasoning |
| ROASBench measured by Spring Prompt · 6 results | ||
| Consistency scorescore out of 100, higher is better 18 · 29 of 33 measured | 28.3 | 14.8 |
| Contribution profitsimulated US dollars, higher is better 15 · 31 of 33 measured | $654,270 | $9,718 |
| Cost of a runUS dollars, lower is better 28 · 12 of 33 measured | $1.01 | $0.13 |
| Months over budgetmonths of 12, lower is better 1 · 1 of 33 measured | 0 | 0 |
| Planning scorescore out of 100, higher is better 20 · 33 of 33 measured | 54.7 | 51.7 |
| Return on ad spendprofit per $1 spent, higher is better 19 · 31 of 33 measured | $0.50 | $0.01 |
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 8 results | ||
| Confirmed task success · AgentIPS effect estimate, higher is better 9 · 16 of 48 measured | 0.08max reasoning, arena agent | 0.05max reasoning, arena agent |
| Creative writing · TextArena rating, higher is better 20 · 28 of 385 measured | 1,460max reasoning | 1,454max reasoning |
| Expert prompts · TextArena rating, higher is better 10 · 16 of 338 measured | 1,533max reasoning | 1,518max reasoning |
| Instruction following · TextArena rating, higher is better 11 · 21 of 387 measured | 1,488max reasoning | 1,476max reasoning |
| Praise over complaint · AgentIPS effect estimate, higher is better 16 · 18 of 48 measured | 0.07max reasoning, arena agent | 0.05max reasoning, arena agent |
| Steerability · AgentIPS effect estimate, higher is better 23 · 26 of 48 measured | 0.01max reasoning, arena agent | 0.01max reasoning, arena agent |
| Tool grounding · AgentIPS effect estimate, higher is better 2 · 12 of 48 measured | 0.00max reasoning, arena agent | 0.00max reasoning, arena agent |
| Writing, literature and language · TextArena rating, higher is better 13 · 23 of 386 measured | 1,476max reasoning | 1,466max reasoning |
| Artificial Analysis reported by Artificial Analysis · 8 results | ||
| Terminal-Bench 2.1% of tasks, higher is better 12 · 17 of 131 measured | 85.0%max reasoning | 83.9%max reasoning |
| Terminal-Bench 4.0% of tasks, higher is better 37 · 13 of 120 measured | 12.6%low reasoning | 41.9%max reasoning |
| τ-bench banking% of tasks, higher is better 10 · 4 of 121 measured | 46.0%max reasoning | 50.3%max reasoning |
| Artificial Analysis Coding Indexindex score, higher is better 10 · 14 of 130 measured | 76.2max reasoning | 74.8max reasoning |
| Long-context reasoning (AA-LCR)% of questions, higher is better 1 · 47 of 241 measured | 88.7%max reasoning | 79.7%max reasoning |
| Output speedtokens per second, higher is better 74 · 49 of 78 measured | 41.0max reasoning | 87.5low reasoning |
| SciCode% of problems, higher is better 11 · 12 of 121 measured | 59.5%max reasoning | 59.0%max reasoning |
| Time to first answer tokenseconds, lower is better 76 · 62 of 78 measured | 51.8max reasoning | 25.4low reasoning |
Compare them with other models
Quick answers
Is Kimi K3 better than GLM-5.3?
Pick Kimi K3 if quality matters most: it scores higher (Spring Prompt overall 47 against 35). Pick GLM-5.3 if cost matters more: it costs 3.5 times less per game of bullet chess on BulletBench. GLM-5.3 is also faster.
Which is cheaper, Kimi K3 or GLM-5.3?
GLM-5.3 is cheaper to run: $0.016 against $0.054 per game of bullet chess on BulletBench (3.5 times less).
Which is faster, Kimi K3 or GLM-5.3?
GLM-5.3 writes about 87 tokens a second on its usual API, against 41 for Kimi K3 (Artificial Analysis).
Which is better for marketing planning, Kimi K3 or GLM-5.3?
Kimi K3 ranks 17 of 32 models on sale for planning a year of ad spend without overspending, against 29 for GLM-5.3.
Which is better for agents and tool use, Kimi K3 or GLM-5.3?
GLM-5.3 ranks 15 of 159 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 40 for Kimi K3.
Which is better for professional work, Kimi K3 or GLM-5.3?
GLM-5.3 ranks 15 of 51 models on sale for real tasks from banking, consulting and law, business documents and freelance projects, against 24 for Kimi K3.
Which is better for reasoning and knowledge, Kimi K3 or GLM-5.3?
Kimi K3 ranks 18 of 187 models on sale for hard questions across science, maths and general knowledge, against 31 for GLM-5.3.
Run Kimi K3 and GLM-5.3 on your own prompt.
Benchmarks aren't your task. Test both side by side on your prompt, with every answer checked against your rules and the cost of each one.