DeepSeek V4 Flash vs GLM 5.3 Flash
Full profiles: DeepSeek V4 Flash (DeepSeek) and GLM 5.3 Flash (Zhipu AI): every benchmark, price and rank.
Updated · Data from 12 sources
As of 16 Nov 2026, DeepSeek V4 Flash and GLM 5.3 Flash are neck and neck on quality (57.2 vs 57.8 on the BenchLeader Index, inside the ±6.2 margin), and they cost about the same at $0.262 vs $0.238 per 1M tokens blended. Choose DeepSeek V4 Flash for reasoning (59 vs 48) and knowledge (55 vs 45) and fast first answers (0.7 s vs 87.0 s); choose GLM 5.3 Flash for maths (69 vs 62), human preference (66 vs 62), and coding (57 vs 55).
Key facts (as of 16 Nov 2026)
- DeepSeek V4 Flash scores 57.2 on the BenchLeader Index and GLM 5.3 Flash scores 57.8 (BenchLeader, 16 Nov 2026).
- DeepSeek V4 Flash costs $0.262 per 1M tokens blended (list price, 16 Nov 2026).
- GLM 5.3 Flash costs $0.238 per 1M tokens blended (list price, 16 Nov 2026).
- Of the 10 benchmarks both models have, GLM 5.3 Flash leads 10 (BenchLeader, 16 Nov 2026).
- On Arena WebDev, DeepSeek V4 Flash is rated 1582 and GLM 5.3 Flash 1609 (LMArena, as published by 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 Flash | 57 vs 55 category score; Arena WebDev 1609 vs 1582 | Medium |
| Maths | GLM 5.3 Flash | 69 vs 62 category score; Arena Maths 1505 vs 1441 | Medium |
| Reasoning | DeepSeek V4 Flash | 59 vs 48 category score; Arena Hard Prompts 1459 vs 1500 | High |
| Knowledge | DeepSeek V4 Flash | 55 vs 45 category score | High |
| Chat quality (human preference) | GLM 5.3 Flash | 66 vs 62 on Arena votes | Medium |
| Lowest cost | Either | $0.262 vs $0.238 per 1M blended; agentic coding session $0.011 vs $0.011 | – |
| Fastest first answer | DeepSeek V4 Flash | 0.7 s (DeepSeek V4 Flash (high)) vs 87.0 s (GLM 5.3 Flash) 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
- DeepSeek V4 Flash (high) and GLM 5.3 Flash are neck and neck on quality (57.2 vs 57.8, within the ±6.2 margin).
- DeepSeek V4 Flash (high) is stronger in knowledge, reasoning.
- GLM 5.3 Flash is stronger in coding, human preference, maths.
- They cost about the same ($0.238 per 1M blended).
- DeepSeek V4 Flash (high) streams 1.5× faster (82 vs 54 tokens per second).
At a glance
| Spec | DeepSeek V4 Flash | GLM 5.3 Flash |
|---|---|---|
| Developer | DeepSeek | Zhipu AI |
| Released | 31 Jul 2026 | 20 Aug 2026 |
| Context window (tokens) | 1,048,576 | 1,000,000 |
| Compared at | DeepSeek V4 Flash (high) | GLM 5.3 Flash |
| Open weights | Yes | Yes |
| Price in / out / cached input per 1M | $0.150 / $0.600 / $0.0030 | $0.150 / $0.500 / $0.030 |
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 | DeepSeek V4 Flash (high) | GLM 5.3 Flash | Leader |
|---|---|---|---|
| Reasoning | 59.2 | 48.3 | DeepSeek V4 Flash |
| Coding | 55.0 | 56.9 | GLM 5.3 Flash |
| Agents & tools | – | 58.7 | not comparable |
| Maths | 61.9 | 68.8 | GLM 5.3 Flash |
| Knowledge | 54.8 | 45.1 | DeepSeek V4 Flash |
| Human preference | 62.0 | 66.4 | GLM 5.3 Flash |
| Multimodal | – | 64.0 | not comparable |
| Long context | – | 64.6 | not comparable |
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: Agents & tools, Instruction following, Long context, Multimodal.
Where sources disagree
Across the 10 benchmarks both have, DeepSeek V4 Flash leads 0, GLM 5.3 Flash leads 10. Every overall index that rates both puts GLM 5.3 Flash ahead.
| Overall index | DeepSeek V4 Flash (high) | GLM 5.3 Flash | Leader |
|---|---|---|---|
| BenchLeader Index | 57.2 | 57.8 | neck and neck |
| Arena Text | 1439 | 1475 | GLM 5.3 Flash |
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 | DeepSeek V4 Flash (high) | GLM 5.3 Flash |
|---|---|---|
| Blended price per 1M tokens | $0.262 | $0.238 |
| Input / output per 1M | $0.150 / $0.600 | $0.150 / $0.500 |
| Cached input per 1M | $0.0030 | $0.030 |
| Price per task: chat reply | $0.0002 | $0.0002 |
| Price per task: summarize a 30-page report | $0.0022 | $0.0021 |
| Price per task: code edit | $0.0018 | $0.0017 |
| Price per task: agentic coding session | $0.011 | $0.011 |
| Price per task: structured extraction | $0.0004 | $0.0004 |
| Agentic coding session, input mostly cached | $0.0048 | $0.0056 |
| Cost per benchmark task (AA Index) | – | $0.583 |
| Output tokens per task (AA Index) | – | 134k |
| Of which reasoning | – | 103k |
| Time per task (AA Index) | – | 41 min |
| Output speed | 82 tok/s | 54 tok/s |
| Time to first answer | 0.7 s | 87.0 s |
| Time to a complete answer, last measured 13 Nov 2026 | – | 220.4 s |
Each model is compared at its best-ranked reasoning-effort setting. DeepSeek V4 Flash (high) scores 57.2 and takes 1 s to its first answer token; DeepSeek V4 Flash (max) scores 56.8 in 28 s, DeepSeek V4 Flash scores 53.8 in 1 s. GLM 5.3 Flash scores 57.8 and takes 87 s to its first answer token; GLM 5.3 Flash (max) scores 52.4 in 87 s. 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
148 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 | DeepSeek V4 Flash (high) | GLM 5.3 Flash |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 57.2 | 57.8 |
| Coding score | BenchLeader | 55.0 | 56.9 |
| Human preference score | BenchLeader | 62.0 | 66.4 |
| Knowledge score | BenchLeader | 54.8 | 45.1 |
| Maths score | BenchLeader | 61.9 | 68.8 |
| Reasoning score | BenchLeader | 59.2 | 48.3 |
| Agents & tools score | BenchLeader | – | 58.7 |
| Long context score | BenchLeader | – | 64.6 |
| Multimodal score | BenchLeader | – | 64.0 |
| Blended price $/M | BenchLeader | $0.262 | $0.238 |
| Output speed | BenchLeader | 82 tok/s | 54 tok/s |
| Time to first answer | BenchLeader | 0.7 s | 87.0 s |
| Context window | BenchLeader | 1,048,576 tokens | 1,000,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 91.0%† | 90.2%† |
| SimpleBench | SimpleBench | 61.1%† | – |
| Arena Hard Prompts | Arena (LMArena) | 1459 | 1500 |
| LiveBench 2.0 Reasoning | LiveBench | – | 77.6%† |
| LiveBench 1.0 Reasoning | LiveBench | 86.6%† | 77.6% |
| GPQA Diamond (AA) | Artificial Analysis | 90.8%† | 91.2% |
| Humanity's Last Exam (AA) | Artificial Analysis | 38.5%† | 39.9% |
| HLE-Diamond (AA) | Artificial Analysis | – | 18.2% |
| GPQA Diamond (Vals) | Vals AI | 89.9% | 86.4%† |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 52.2%† | – |
| ARC-AGI-1 | ARC Prize | 87.0% | 91.0%† |
| ARC-AGI-2 | ARC Prize | 56.0% | 65.8%† |
| CritPt | Artificial Analysis | 16.6%† | 15.4% |
| CritPt v2 hard subset (AA) | Artificial Analysis | – | 24.5% |
| Arena Expert | Arena (LMArena) | 1467 | 1511 |
| Chess Puzzles | Epoch AI Benchmarking Hub | 32.0%† | 21.0%† |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 34.0%† | 8.0%† |
| LMCA | Epoch AI Benchmarking Hub | 41.7%† | – |
| DTBench | Epoch AI Benchmarking Hub | 90.9%† | – |
| BTF-3 | Epoch AI Benchmarking Hub | – | 14.9% |
| Coding | |||
| SciCode | SciCode | 42.0% | 51.6% |
| WeirdML | WeirdML | 57.0% | – |
| FrontierCode | Cognition | 18.8%† | 31.8%† |
| Arena Coding | Arena (LMArena) | 1480 | 1526 |
| Arena WebDev | Arena (LMArena) | 1582 | 1609 |
Show all 148 benchmarks
| Metric | Publisher | DeepSeek V4 Flash (high) | GLM 5.3 Flash |
|---|---|---|---|
| Coding | |||
| LiveBench 2.0 Coding | LiveBench | – | 70.9%† |
| LiveBench 1.0 Coding | LiveBench | 75.0%† | 79.0% |
| SciCode (AA) | Artificial Analysis | 50.4%† | 51.6% |
| LiveCodeBench | Vals AI | 87.3% | 80.5%† |
| IOI | Vals AI | 32.7% | 52.5%† |
| SWE-bench (Vals) | Vals AI | 88.8% | 92.0%† |
| Code Migration | Vals AI | 38.6% | 20.5%† |
| Code Migration v1.1 | Vals AI | – | 51.6%† |
| ProgramBench | Vals AI | 0.0% | 0.0%† |
| Vibe Code Bench 1-100 | Vals AI | – | 16.0%† |
| Vibe Code Bench v1.1 | Vals AI | 74.7% | 30.8%† |
| SWE Atlas v1.1 | Scale AI SEAL | – | 55.3% |
| SWE Atlas v1.1: Codebase QnA | Scale AI SEAL | – | 66.9% |
| SWE Atlas v1.1: Test Writing | Scale AI SEAL | – | 53.3% |
| SWE Atlas v1.1: Refactoring | Scale AI SEAL | – | 45.7% |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | – | 52.5%† |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | – | 63.4%† |
| CursorBench | Cursor | – | 36.8%† |
| ALE-Bench | Epoch AI Benchmarking Hub | 678.2 | 303.6† |
| FrontierSWE | Epoch AI Benchmarking Hub | – | 18.1%† |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 12.1%† | 32.8% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 78.7%† | 84.3% |
| GameBench 2 Overall | DuelLab GameBench 2 | 34.1† | 54.9 |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 43.4 | 57.2† |
| Design Arena Website | Design Arena | 1199† | 1257 |
| FrontierCode 1.1 (AA) | Artificial Analysis | – | 41.0% |
| FrontierCode 2.0 (AA) | Artificial Analysis | – | 17.6% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | – | 37.0%† |
| APEX-Agents | Mercor | – | 52.8% |
| Arena Agent | Arena (LMArena) | – | -2.9 |
| LiveBench Long Agentic | LiveBench | – | 20.7%† |
| LiveBench Browser | LiveBench | – | 59.8%† |
| LiveBench Agentic Coding | LiveBench | 46.8%† | 56.8% |
| Terminal-Bench 2.1 (Vals) | Vals AI | 67.0%† | 62.9%† |
| HiL-Bench | Scale AI SEAL | 47.8%† | – |
| GDPval-AA v2.1, normalised score | Artificial Analysis | – | 57.1% |
| GDPval-AA v2.1 | Artificial Analysis | 1437† | 1648 |
| τ³-Banking (AA) | Artificial Analysis | 39.4%† | 47.2% |
| ITBench SRE (AA) | Artificial Analysis | – | 51.2% |
| Analyst Agent (AA) | Artificial Analysis | – | 25.0% |
| CyberBench | Vals AI | 72.1% | 67.9%† |
| Finance Agent v2 | Vals AI | 49.5% | 57.9%† |
| Finance Agent v2.1 | Vals AI | – | 60.6%† |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | – | 10.4%† |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 8.3% | 6.7%† |
| Legal Research Bench | Vals AI | 30.3% | 45.2%† |
| SkillsBench | Vals AI | 50.7% | 40.2%† |
| SREBench | Vals AI | – | 0.8%† |
| Tax Agent Bench | Vals AI | 28.1% | 31.6%† |
| Terminal-Bench 4.0 (Vals) | Vals AI | 18.7% | 37.4%† |
| Terminal-Bench Science | Vals AI | 4.3% | 2.9%† |
| Arena Agent Task Completion | Arena (LMArena) | – | 1.3 |
| PostTrainBench | Epoch AI Benchmarking Hub | – | 36.8%† |
| AutomationBench | Artificial Analysis | 54.0%† | 60.4% |
| GDP.pdf (AA) | Artificial Analysis | 11.0%† | 15.4% |
| Harvey LAB-AA v1.1 | Harvey | – | 0.8% |
| EnterpriseOps-Gym | Artificial Analysis | – | 33.2% |
| AA-Briefcase v1.1 | Artificial Analysis | 1255† | 1450 |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | – | 4.8% |
| CWE-Bench (AA) | Artificial Analysis | – | 55.8% |
| DeepSec-Bench (AA) | Artificial Analysis | – | 19.9% |
| CyberGym end-to-end (AA) | Artificial Analysis | – | 74.0% |
| MKA (AA) | Artificial Analysis | – | 36.2% |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 57.5%† | 55.8%† |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 24.4%† | 17.1%† |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 94.4%† | 93.9%† |
| ProofBench | Vals AI | 56.0%† | 21.0%† |
| LiveBench Mathematics | LiveBench | 86.8%† | 81.2% |
| AIME 2026 | MathArena | 95.8%† | – |
| HMMT February 2026 | MathArena | 93.9%† | – |
| MathArena Apex | MathArena | 27.1%† | – |
| Arena Maths | Arena (LMArena) | 1441 | 1505 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 33.6%† | – |
| LiveBench 2.0 Data Analysis | LiveBench | – | 76.4%† |
| LiveBench Research | LiveBench | – | 73.9%† |
| LiveBench 1.0 Data Analysis | LiveBench | 79.3%† | 76.4% |
| AA-Omniscience v1.1 | Artificial Analysis | – | 0.7 |
| MMLU-Pro | Vals AI | 86.2% | 86.1%† |
| LegalBench | Vals AI | 77.7% | 83.9%† |
| CorpFin | Vals AI | 61.9% | – |
| TaxEval | Vals AI | 70.7% | 75.6%† |
| BioMysteryBench | Vals AI | 64.4% | – |
| Excel Modeling Benchmark | Vals AI | 57.0% | 55.9%† |
| Excel Modeling Benchmark v1.1 | Vals AI | – | 66.3%† |
| MedCode | Vals AI | 41.4% | – |
| MedScribe | Vals AI | 80.4% | 88.9%† |
| Access to Justice Benchmark | Vals AI | – | 71.1%† |
| GDP.pdf | Epoch AI Benchmarking Hub | – | 14.0%† |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | – | 28.2% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | – | 61.6% |
| Instruction following | |||
| LiveBench Language | LiveBench | 79.2%† | 77.3% |
| LiveBench Instruction Following | LiveBench | 65.5%† | 52.8% |
| Arena Instruction Following | Arena (LMArena) | 1434 | 1474 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1439 | 1475 |
| EQ-Bench Creative Writing v3 | EQ-Bench | 1559† | – |
| Arena Creative Writing | Arena (LMArena) | 1411 | 1435 |
| Arena Multi-turn | Arena (LMArena) | 1447 | 1476 |
| Arena Longer Queries | Arena (LMArena) | 1449 | 1480 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1295 |
| LiveBench Design | LiveBench | – | 73.7%† |
| MMMU-Pro (Vals) | Vals AI | – | 86.0%† |
| SAGE | Vals AI | – | 89.3%† |
| Humanity's Sixth Sense | Scale AI SEAL | – | 26.8%† |
| VIALS-AA | Handshake AI | – | 39.8% |
| Code2Video-AA | Artificial Analysis | – | 975 |
| AA-3D | Artificial Analysis | – | 937 |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 79.7%† | 80.0% |
| MLCR | Artificial Analysis | – | 11.7% |
| FACTS Grounding | – | 56.1% | |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 154.4† | 153.8 |
| LiveBench 2.0 | LiveBench | – | 60.7%† |
| LiveBench 1.0 | LiveBench | 74.2%† | 71.6% |
| AA Intelligence Index v5.0.1 | Artificial Analysis | – | 27.8 |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 34.3† | 41.8 |
| AA Progress Index v1 | Artificial Analysis | 309.4† | 318.6 |
| Vals Index | Vals AI | – | 50.9† |
| Vals Finance Index | Vals AI | – | 55.8%† |
| Vals Router Index | Vals AI | – | 50.9% |
| Vals Legal Index | Vals AI | – | 71.0%† |
| AA Cyber Index | Artificial Analysis | – | 49.9 |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | – | 26.7% |
| DistressBench | Scale AI SEAL | – | 72.7% |
| Metadata | |||
| AA Openness Index | Artificial Analysis | 44.4† | 44.4 |
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
- GLM 5.3 Flash: BenchLeader Index 56.7 → 57.8.
- DeepSeek V4 Flash (high): 1 score revision.
- GLM 5.3 Flash: 1 score revision.
- GLM 5.3 Flash: 5 score revisions.
- GLM 5.3 Flash: 1 new result (FORTRESS).
- DeepSeek V4 Flash (high): 1 score revision.
- GLM 5.3 Flash: 2 score revisions; GLM 5.3 Flash cached input price moved from $0.034/M to $0.030/M (13% down).
- GLM 5.3 Flash: GLM 5.3 Flash cached input price moved from $0.030/M to $0.034/M (15% up).
- DeepSeek V4 Flash (high): 1 new result (GameBench 2 by reasoning setting).
- GLM 5.3 Flash: GLM 5.3 Flash cached input price moved from $0.034/M to $0.030/M (13% down).
Related comparisons
- DeepSeek V4 Flash vs Claude Fable 5.1
- DeepSeek V4 Flash vs GPT-6.1 Sol
- GLM 5.3 Flash vs Claude Opus 5.5
- GLM 5.3 Flash vs GPT-6.1 Sol
- GLM 5.3 Flash vs Gemini 4 Argon
- DeepSeek V4 Flash vs GLM 5.3
- DeepSeek V4 Flash vs Gemini 4 Argon
- DeepSeek V4 Flash vs Claude Sonnet 5.5
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
DeepSeek V4 Flash vs GLM 5.3 Flash: questions
- Is DeepSeek V4 Flash better than GLM 5.3 Flash?
- DeepSeek V4 Flash (high) and GLM 5.3 Flash are neck and neck on quality (57.2 vs 57.8, within the ±6.2 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.
- DeepSeek V4 Flash vs GLM 5.3 Flash: which is cheaper, and will that change?
- GLM 5.3 Flash is cheaper today: $0.238 against $0.262 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is DeepSeek V4 Flash or GLM 5.3 Flash better for coding?
- GLM 5.3 Flash scores higher in coding (57 vs 55 on the category index, where 50 is average). On Arena WebDev, the top shared coding benchmark, it is 1582 for DeepSeek V4 Flash against 1609 for GLM 5.3 Flash.
- Which is faster, DeepSeek V4 Flash or GLM 5.3 Flash?
- DeepSeek V4 Flash streams faster (82 vs 54 output tokens per second, Artificial Analysis 7-day median).
- Which is better for long documents?
- Both take about 1M tokens: DeepSeek V4 Flash 1,048,576, GLM 5.3 Flash 1,000,000.
More questions (3)
- Which has lower latency?
- DeepSeek V4 Flash answers first: 0.7 s against 87.0 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 DeepSeek V4 Flash vs GLM 5.3 Flash?
- Across the 10 benchmarks both have, DeepSeek V4 Flash leads 0, GLM 5.3 Flash leads 10. Every overall index that rates both puts GLM 5.3 Flash ahead.
- When was GLM 5.3 Flash released?
- GLM 5.3 Flash was released on 20 Aug 2026.