GLM 5.3 vs Step 5 Preview
Full profiles: GLM 5.3 (Zhipu AI) and Step 5 Preview (StepFun): every benchmark, price and rank.
Updated · Data from 10 sources
As of 16 Nov 2026, GLM 5.3 and Step 5 Preview are neck and neck on quality (58.8 vs 56.0 on the BenchLeader Index, inside the ±5.3 margin), and Step 5 Preview is 1.5× cheaper at $1.43 vs $2.15 per 1M tokens blended. Choose GLM 5.3 for coding (60 vs 50), agents & tools (58 vs 49), and human preference (67 vs 63); choose Step 5 Preview for fast first answers (23.5 s vs 57.6 s) and long context (69 vs 65), where it also streams ~2× faster.
Key facts (as of 16 Nov 2026)
- GLM 5.3 scores 58.8 on the BenchLeader Index and Step 5 Preview scores 56.0 (BenchLeader, 16 Nov 2026).
- GLM 5.3 costs $2.15 per 1M tokens blended (list price, 16 Nov 2026).
- Step 5 Preview costs $1.43 per 1M tokens blended (list price, 16 Nov 2026).
- Step 5 Preview streams ~2× faster than GLM 5.3 (Artificial Analysis, 7-day median, 16 Nov 2026).
- Of the 60 benchmarks both models have, GLM 5.3 leads 42 (BenchLeader, 16 Nov 2026).
Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | GLM 5.3 | 60 vs 50 category score; FrontierCode 2.0 (AA) 21.6% vs 12.0% | High |
| Agents & tool use | GLM 5.3 | 58 vs 49 category score; Arena Agent 1 vs -3.7 | High |
| Maths | Either | 60 vs 60 category score; Arena Maths 1501 vs 1470 | Either |
| Reasoning | GLM 5.3 | 55 vs 52 category score; CritPt v2 hard subset (AA) 33.5% vs 32.5% | Medium |
| Knowledge | Either | 54 vs 54 category score; AA-Omniscience v1.1 6.0 vs 16.4 | Either |
| Long documents | Step 5 Preview | 69 vs 65 category score; AA-LCR v1.1 88.3% vs 79.7% | Medium |
| Chat quality (human preference) | GLM 5.3 | 67 vs 63 on Arena votes | Medium |
| Lowest cost | Step 5 Preview | $1.43 vs $2.15 per 1M blended; agentic coding session $0.071 vs $0.102 | – |
| Fastest first answer | Step 5 Preview | 23.5 s (Step 5 Preview) vs 57.6 s (GLM 5.3 (max)) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- GLM 5.3 (max) and Step 5 Preview are neck and neck on quality (58.8 vs 56.0, within the ±5.3 margin).
- GLM 5.3 (max) is stronger in agents & tools, coding, human preference, reasoning.
- Step 5 Preview is stronger in knowledge, long context, maths.
- Step 5 Preview is 1.5× cheaper ($1.43 vs $2.15 per 1M blended).
- Step 5 Preview streams 1.4× faster (124 vs 88 tokens per second).
At a glance
| Spec | GLM 5.3 | Step 5 Preview |
|---|---|---|
| Developer | Zhipu AI | StepFun |
| Released | 18 Aug 2026 | 18 Sep 2026 |
| Context window (tokens) | 1,000,000 | 1,000,000 |
| Compared at | GLM 5.3 (max) | Step 5 Preview |
| Open weights | Yes | Yes |
| Price in / out / cached input per 1M | $1.40 / $4.40 / $0.260 | $1.00 / $2.70 / $0.105 |
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | GLM 5.3 (max) | Step 5 Preview | Leader |
|---|---|---|---|
| Reasoning | 54.7 | 51.7 | GLM 5.3 |
| Coding | 59.7 | 49.9 | GLM 5.3 |
| Agents & tools | 57.6 | 49.1 | GLM 5.3 |
| Maths | 60.0 | 60.1 | level |
| Knowledge | 53.5 | 53.7 | level |
| Human preference | 67.2 | 63.1 | GLM 5.3 |
| Multimodal | – | 59.3 | not comparable |
| Long context | 64.5 | 69.0 | Step 5 Preview |
On one chart
Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.
Not drawn, measured for fewer than two of these models: Instruction following, Multimodal.
Where sources disagree
Across the 60 benchmarks both have, GLM 5.3 leads 42, Step 5 Preview leads 17 and 1 is a tie. Every overall index that rates both puts GLM 5.3 ahead.
| Overall index | GLM 5.3 (max) | Step 5 Preview | Leader |
|---|---|---|---|
| BenchLeader Index | 58.8 | 56.0 | neck and neck |
| AA Intelligence Index v5.0.1 | 31.6 | 28.5 | GLM 5.3 |
| Vals Index | 59.9 | 54.8 | GLM 5.3 |
| Arena Text | 1482 | 1449 | GLM 5.3 |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.
| Measure | GLM 5.3 (max) | Step 5 Preview |
|---|---|---|
| Blended price per 1M tokens | $2.15 | $1.43 |
| Input / output per 1M | $1.40 / $4.40 | $1.00 / $2.70 |
| Cached input per 1M | $0.260 | $0.105 |
| Price per task: chat reply | $0.0019 | $0.0012 |
| Price per task: summarize a 30-page report | $0.019 | $0.014 |
| Price per task: code edit | $0.015 | $0.010 |
| Price per task: agentic coding session | $0.102 | $0.071 |
| Price per task: structured extraction | $0.0037 | $0.0025 |
| Agentic coding session, input mostly cached | $0.050 | $0.030 |
| Cost per benchmark task (AA Index) | $5.01 | $2.35 |
| Output tokens per task (AA Index) | 119k | 113k |
| Of which reasoning | 85k | 80k |
| Time per task (AA Index) | 23 min | 15 min |
| Output speed | 88 tok/s | 124 tok/s |
| Time to first answer | 57.6 s | 23.5 s |
| Time to a complete answer, last measured 13 Nov 2026 | 144.1 s | 54.4 s |
Per token is not per task: an average task of Artificial Analysis' Intelligence Index cost $5.01 on GLM 5.3 (max) and $2.35 on Step 5 Preview, and GLM 5.3 (max) wrote 1.0× as many output tokens for the same work.
Each model is compared at its best-ranked reasoning-effort setting. GLM 5.3 (max) scores 58.8 and takes 58 s to its first answer token; GLM 5.3 (low) scores 47.3 in 7 s. Step 5 Preview scores 56.0 and takes 24 s to its first answer token. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
155 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | GLM 5.3 (max) | Step 5 Preview |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 58.8 | 56.0 |
| Agents & tools score | BenchLeader | 57.6 | 49.1 |
| Coding score | BenchLeader | 59.7 | 49.9 |
| Human preference score | BenchLeader | 67.2 | 63.1 |
| Knowledge score | BenchLeader | 53.5 | 53.7 |
| Long context score | BenchLeader | 64.5 | 69.0 |
| Maths score | BenchLeader | 60.0 | 60.1 |
| Reasoning score | BenchLeader | 54.7 | 51.7 |
| Multimodal score | BenchLeader | – | 59.3 |
| Blended price $/M | BenchLeader | $2.15 | $1.43 |
| Output speed | BenchLeader | 88 tok/s | 124 tok/s |
| Time to first answer | BenchLeader | 57.6 s | 23.5 s |
| Context window | BenchLeader | 1,000,000 tokens | 1,000,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 90.9% | – |
| SimpleBench | SimpleBench | 66.2%† | – |
| Arena Hard Prompts | Arena (LMArena) | 1507 | 1477 |
| LiveBench 2.0 Reasoning | LiveBench | – | 83.8%† |
| LiveBench 1.0 Reasoning | LiveBench | 85.8%† | – |
| GPQA Diamond (AA) | Artificial Analysis | 91.7% | – |
| Humanity's Last Exam (AA) | Artificial Analysis | 42.3% | 46.5% |
| HLE-Diamond (AA) | Artificial Analysis | 17.9% | 24.3% |
| GPQA Diamond (Vals) | Vals AI | 88.1% | – |
| CritPt | Artificial Analysis | 19.1% | 20.9% |
| CritPt v2 hard subset (AA) | Artificial Analysis | 33.5% | 32.5% |
| MysteryMechanism | Vals AI | 23.0% | 18.5% |
| HLE-Diamond (Scale) | Scale AI SEAL | 16.4%† | – |
| Arena Expert | Arena (LMArena) | 1522 | 1489 |
| Chess Puzzles | Epoch AI Benchmarking Hub | 37.0% | – |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 33.0% | – |
| LMCA | Epoch AI Benchmarking Hub | 55.5%† | – |
| DTBench | Epoch AI Benchmarking Hub | 87.7%† | – |
| BTF-3 | Epoch AI Benchmarking Hub | 14.9%† | – |
| Coding | |||
| SciCode | SciCode | 59.0% | 58.9% |
| WeirdML | WeirdML | 75.4% | – |
| FrontierCode | Cognition | 40.1% | – |
| Arena Coding | Arena (LMArena) | 1522 | 1505 |
| Arena WebDev | Arena (LMArena) | 1623 | 1573† |
| LiveBench 2.0 Coding | LiveBench | – | 67.6%† |
Show all 155 benchmarks
| Metric | Publisher | GLM 5.3 (max) | Step 5 Preview |
|---|---|---|---|
| Coding | |||
| LiveBench 1.0 Coding | LiveBench | 79.0%† | – |
| LiveCodeBench | Vals AI | 80.5% | – |
| IOI | Vals AI | 68.4% | 38.1% |
| SWE-bench (Vals) | Vals AI | 95.4% | – |
| Code Migration | Vals AI | 44.2% | 39.0% |
| Code Migration v1.1 | Vals AI | 55.1% | 47.0% |
| ProgramBench | Vals AI | 1.5% | – |
| SWE-bench Multimodal | Vals AI | – | 40.8% |
| Vibe Code Bench 1-100 | Vals AI | 20.0% | – |
| Vibe Code Bench v1.1 | Vals AI | 78.1% | 70.4% |
| Vibe Code Bench 1-100 v1.1 | Vals AI | 29.8% | – |
| SWE Atlas v1.1 | Scale AI SEAL | 58.1%† | – |
| SWE Atlas v1.1: Codebase QnA | Scale AI SEAL | 70.2%† | – |
| SWE Atlas v1.1: Test Writing | Scale AI SEAL | 55.6%† | – |
| SWE Atlas v1.1: Refactoring | Scale AI SEAL | 48.6%† | – |
| SWE-Bench Pro V2 | Scale AI SEAL | 95.6% | – |
| SWE-Bench Pro V2 HARD | Scale AI SEAL | 84.3% | – |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 69.0% | – |
| CursorBench | Cursor | 42.6% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 1317.4† | – |
| FrontierSWE | Epoch AI Benchmarking Hub | 30.2% | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 41.9% | 33.3% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 83.9% | – |
| Startrise Frontend | Startrise | 73.2%† | – |
| GameBench 2 Overall | DuelLab GameBench 2 | 64.6† | – |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 72.3 | – |
| Design Arena Website | Design Arena | 1279† | – |
| FrontierCode 1.1 (AA) | Artificial Analysis | 40.1% | 38.4% |
| FrontierCode 2.0 (AA) | Artificial Analysis | 21.6% | 12.0% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 41.2% | – |
| APEX-Agents | Mercor | 56.6%† | – |
| Arena Agent | Arena (LMArena) | 1 | -3.7 |
| LiveBench Long Agentic | LiveBench | – | 19.4%† |
| LiveBench Browser | LiveBench | – | 24.8%† |
| LiveBench Agentic Coding | LiveBench | 60.9%† | – |
| Terminal-Bench 2.1 (Vals) | Vals AI | 71.5% | – |
| MCP Atlas | Scale AI SEAL | 84.2%† | – |
| HiL-Bench | Scale AI SEAL | 58.3%† | – |
| GDPval-AA v2.1, normalised score | Artificial Analysis | 57.5% | 54.3% |
| GDPval-AA v2.1 | Artificial Analysis | 1655 | 1591 |
| τ³-Banking (AA) | Artificial Analysis | 50.3% | – |
| ITBench SRE (AA) | Artificial Analysis | 46.1% | 55.6% |
| Analyst Agent (AA) | Artificial Analysis | 38.8% | 35.0% |
| APEX-Agents (AA) | Artificial Analysis | – | 38.0% |
| CyberBench | Vals AI | 72.1% | – |
| Finance Agent v2 | Vals AI | 55.8% | 50.7% |
| Finance Agent v2.1 | Vals AI | 65.0% | 58.9% |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | 14.6% | 12.9% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 8.3% | 9.2% |
| Legal Research Bench | Vals AI | 49.0% | 40.4% |
| SkillsBench | Vals AI | 47.5% | – |
| SREBench | Vals AI | 9.2% | 0.4% |
| Tax Agent Bench | Vals AI | 43.3% | 27.3% |
| Terminal-Bench 4.0 (Vals) | Vals AI | 38.9% | 31.8% |
| Terminal-Bench Science | Vals AI | 5.7% | 2.9% |
| Arena Agent Task Completion | Arena (LMArena) | 3.1 | -1.5 |
| PostTrainBench | Epoch AI Benchmarking Hub | 34.5% | – |
| Vending-Bench 2 | Andon Labs | 8163.6† | – |
| AutomationBench | Artificial Analysis | 62.2% | 51.0% |
| GDP.pdf (AA) | Artificial Analysis | 11.2% | 14.8% |
| Harvey LAB-AA v1.1 | Harvey | 0.3% | – |
| EnterpriseOps-Gym | Artificial Analysis | 36.4% | 47.2% |
| AA-Briefcase v1.1 | Artificial Analysis | 1501 | 1424 |
| GDPx | GDPx by SemiAnalysis | 2.7% | – |
| Design Arena Agentic | Design Arena | 1239† | – |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 9.5% | 2.4% |
| CWE-Bench (AA) | Artificial Analysis | 55.8% | 52.5% |
| DeepSec-Bench (AA) | Artificial Analysis | 24.3% | 15.8% |
| CyberGym end-to-end (AA) | Artificial Analysis | 29.0% | 74.8% |
| MKA (AA) | Artificial Analysis | – | 27.6% |
| FACTS Search | 87.7%† | – | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 68.8% | – |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 29.3% | – |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 91.1% | – |
| ProofBench | Vals AI | 49.0% | 42.0% |
| LiveBench Mathematics | LiveBench | 87.9%† | – |
| Arena Maths | Arena (LMArena) | 1501 | 1470 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 41.0% | – |
| LiveBench 2.0 Data Analysis | LiveBench | – | 63.5%† |
| LiveBench Research | LiveBench | – | 75.5%† |
| LiveBench 1.0 Data Analysis | LiveBench | 70.2%† | – |
| AA-Omniscience v1.1 | Artificial Analysis | 6.0 | 16.4 |
| MMLU-Pro | Vals AI | 86.8% | – |
| LegalBench | Vals AI | 84.8% | – |
| TaxEval | Vals AI | 72.4% | – |
| BioMysteryBench | Vals AI | – | 67.4% |
| Excel Modeling Benchmark | Vals AI | 56.3% | 57.1% |
| Excel Modeling Benchmark v1.1 | Vals AI | 69.1% | 69.5% |
| MedCode | Vals AI | 42.9% | – |
| MedScribe | Vals AI | 88.8% | – |
| Public Benefits Bench v1.1 | Vals AI | 68.5% | 58.1% |
| Access to Justice Benchmark | Vals AI | 73.7% | – |
| Financial Coaching Bench | Vals AI | 74.1% | – |
| Immigration Bench | Vals AI | 66.7% | – |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | 32.3% | 36.4% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | 61.1% | 68.7% |
| HealthBench Professional (AA) | OpenAI | 52.8% | – |
| FACTS Parametric | 40.3%† | – | |
| Instruction following | |||
| LiveBench Language | LiveBench | 79.9%† | – |
| LiveBench Instruction Following | LiveBench | 69.3%† | – |
| Arena Instruction Following | Arena (LMArena) | 1481 | 1450 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1482 | 1449 |
| EQ-Bench Creative Writing v3 | EQ-Bench | 2076† | – |
| Arena Creative Writing | Arena (LMArena) | 1456 | 1413 |
| Arena Multi-turn | Arena (LMArena) | 1482 | 1445 |
| Arena Longer Queries | Arena (LMArena) | 1488 | 1461 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1263 |
| LiveBench Design | LiveBench | – | 70.8%† |
| MMMU-Pro | Artificial Analysis | – | 76.4% |
| Code2Video-AA | Artificial Analysis | 1034 | 821 |
| AA-3D | Artificial Analysis | 794 | 859 |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 79.7% | 88.3% |
| MLCR | Artificial Analysis | 17.2% | 18.3% |
| FACTS Grounding | 55.2%† | – | |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 157.1† | – |
| LiveBench 2.0 | LiveBench | – | 54.9%† |
| LiveBench 1.0 | LiveBench | 76.1%† | – |
| AA Intelligence Index v5.0.1 | Artificial Analysis | 31.6 | 28.5 |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 44.8 | 43.7 |
| AA Progress Index v1 | Artificial Analysis | 322.7 | 318.2 |
| Vals Index | Vals AI | 59.9 | 54.8 |
| Vals Finance Index | Vals AI | 61.4% | 55.4% |
| Vals RSI Index | Vals AI | 22.9% | – |
| Vals Science Index | Vals AI | – | 29.6% |
| Vals Router Index | Vals AI | 59.7%† | – |
| Vals Legal Index | Vals AI | 74.6% | 70.2% |
| AA Cyber Index | Artificial Analysis | 36.4 | 47.7 |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | 28.2% | – |
| DistressBench | Scale AI SEAL | 72.5%† | – |
| Metadata | |||
| AA Openness Index | Artificial Analysis | 33.3 | 33.3 |
Data as of 2026-11-16. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261111-7b598bb13de4 (https://benchmarks.duellab.org/releases/gb2-20261111-7b598bb13de4/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- Step 5 Preview: BenchLeader Index 54.2 → 56.0.
- Step 5 Preview: BenchLeader Index 54.7 → 54.2.
- GLM 5.3 (max): 1 new result (PostTrainBench).
- Step 5 Preview: 1 score revision.
- GLM 5.3 (max): 4 score revisions.
- Step 5 Preview: 7 score revisions.
- GLM 5.3 (max): 1 score revision; cached input price moved from $0.30/M to $0.26/M (13% down).
- GLM 5.3 (max): cached input price moved from $0.26/M to $0.30/M (15% up).
- GLM 5.3 (max): cached input price moved from $0.30/M to $0.26/M (13% down).
- GLM 5.3 (max): cached input price moved from $0.26/M to $0.30/M (15% up).
Related comparisons
- GLM 5.3 vs Gemini 4 Carbon
- GLM 5.3 vs Claude Opus 5.5
- GLM 5.3 vs GPT-6 Astra
- Step 5 Preview vs Claude Opus 5.5
- Step 5 Preview vs GPT-6 Astra
- Step 5 Preview vs GPT-6.1 Sol
- Step 5 Preview vs Gemini 4 Argon
- GLM 5.2 vs Step 5 Preview
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
GLM 5.3 vs Step 5 Preview: questions
- Is GLM 5.3 better than Step 5 Preview?
- GLM 5.3 (max) and Step 5 Preview are neck and neck on quality (58.8 vs 56.0, within the ±5.3 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 16 Nov 2026, so check the category scores for your use.
- GLM 5.3 vs Step 5 Preview: which is cheaper, and will that change?
- Step 5 Preview is cheaper today: $1.43 against $2.15 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is GLM 5.3 or Step 5 Preview better for coding?
- GLM 5.3 scores higher in coding (60 vs 50 on the category index, where 50 is average). On FrontierCode 2.0 (AA), the top shared coding benchmark, it is 21.6% for GLM 5.3 against 12.0% for Step 5 Preview.
- Which is faster, GLM 5.3 or Step 5 Preview?
- Step 5 Preview streams ~2× faster (124 vs 88 output tokens per second, Artificial Analysis 7-day median).
- Is GLM 5.3 or Step 5 Preview better for agents and tool use?
- GLM 5.3 scores higher in agentic tasks (58 vs 49 on the category index, where 50 is average). On Arena Agent, the top shared agents & tools benchmark, it is 1 for GLM 5.3 against -3.7 for Step 5 Preview.
More questions (4)
- Which is better for long documents?
- Step 5 Preview scores higher on long-context benchmarks (69 vs 65). Both accept 1,000,000 tokens of context.
- Which has lower latency?
- Step 5 Preview answers first: 23.5 s against 57.6 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Do independent benchmarks agree on GLM 5.3 vs Step 5 Preview?
- Across the 60 benchmarks both have, GLM 5.3 leads 42, Step 5 Preview leads 17 and 1 is a tie. Every overall index that rates both puts GLM 5.3 ahead.
- When was Step 5 Preview released?
- Step 5 Preview was released on 18 Sep 2026.