GLM-5.3-Flash (max): benchmark scores, price and ranking

GLM-5.3-Flash (max) from Z.ai (Zhipu AI) ranks #248 of 743 LLMs on the LLMs Tiger aggregate benchmark score (49.6), based on 14 benchmark results captured up to 2026-10-13.

Organisation

Z.ai (Zhipu AI)

Type

Open-weight / local model

Open weights

Yes

Released

2026-08-20 (source: epoch)

Cheapest API price

$0.110 in / $0.350 out per million tokens (via sail, observed 2026-10-13T19:47Z)

Aggregate rank

#248 of 743 (score 49.6)

Relative to other models, GLM-5.3-Flash (max) is strongest on GPQA Diamond (#35 of 275) and weakest on GDP.pdf (#21 of 25).

Across the 14 benchmarks where at least ten models are scored, GLM-5.3-Flash (max) is in the top 10% on 0 and in the top half on 7.

Its cheapest listed input price, $0.110 per million tokens, is lower than that of 79% of the 427 models with a listed price.

GLM-5.3-Flash (max) benchmark results

About GLM-5.3-Flash (max)

GLM-5.3-Flash (max) is an open-weight model that can run on your own hardware from Z.ai (Zhipu AI). It was released on 2026-08-20, according to Epoch AI. The cheapest API listing we track charges $0.110 per million input tokens and $0.350 per million output tokens, through sail. We hold 14 benchmark results for it across 4 areas, and none is in the top tenth of its board.

Where GLM-5.3-Flash (max) stands

On our aggregate score, GLM-5.3-Flash (max) is #248 of 743 ranked models, around the middle of the table. Its score is 49.6; the scale is centred on 50, so it sits 0.4 points below the typical model. That beats 67% of the ranked field. Among the 6 ranked Z.ai (Zhipu AI) models it is #4. The next one up is GLM-5.1 at 49.6. The next one down is GLM-4.7 (Novita) at 48.4.

GLM-5.3-Flash (max), benchmark by benchmark

GPQA Diamond puts GLM-5.3-Flash (max) at 90.2%, #35 of 275. That beats 87% of the models tested, but 5.6 points short of GPT-6 Astra (max) (95.8%). The results either side are Hy3 (90.4%) and deepseek-v4.1-flash-max (89.8%). It covers expert-level science questions. The board median is 71.7%, so GLM-5.3-Flash (max) is 18.4 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On OTIS Mock AIME, GLM-5.3-Flash (max) reaches 93.9%. It ranks #30 of 189, ahead of 84% of the field, and 6.1 points behind Claude Fable 5.1 (max) at 100.0%. The results either side are DeepSeek V4 Flash 0731 (max) (94.4%) and GLM-5.1 (93.3%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so GLM-5.3-Flash (max) is 36.1 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GLM-5.3-Flash (max) manages 65.8% on ARC-AGI-2, good for #24 of 87. GPT-6 Astra (max) leads at 95.0%, 29.2 points further up, and 72% of models sit below. The results either side are Claude Opus 4.6 (120K, Medium) (66.2%) and Grok 4.20 (Reasoning) (65.1%). It covers abstract reasoning, ARC Prize public eval set. The board median is 22.8%, so GLM-5.3-Flash (max) is 43.1 points above it. Source: ARC Prize official leaderboard, captured 2026-10-13.

Surface Evolver puts GLM-5.3-Flash (max) at 52.5%, #5 of 13. That beats 62% of the models tested, but 40.5 points short of Kimi K3 (max) (93.0%). The results either side are MiniMax-M3 (55.0%) and Kimi K2.7 Code (48.8%). It covers geometry optimisation with a solver. The board median is 46.2%, so GLM-5.3-Flash (max) is 6.2 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On Chess Puzzles, GLM-5.3-Flash (max) reaches 14.0%. It ranks #57 of 122, ahead of 53% of the field, and 58.0 points behind GPT-6 Astra (max) at 72.0%. The results either side are Claude Opus 4.6 (max) (14.0%) and MiniMax-M3 (14.0%). It covers chess tactics puzzles. The board median is 12.5%, so GLM-5.3-Flash (max) is 1.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GLM-5.3-Flash (max) manages 55.8% on FrontierMath T1-3, good for #34 of 68. GPT-6 Astra (max) leads at 93.7%, 37.9 points further up, and 50% of models sit below. The results either side are Kimi K2.6 (57.2%) and GPT-5 (high) (55.4%). It covers research-level maths, tiers 1-3. The board median is 55.6%, so GLM-5.3-Flash (max) is 0.2 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GLM-5.3-Flash (max) lands below the middle on FrontierSWE: 18.1%, #15 of 19, with 74% of the field scoring higher. GPT-6 Astra (max) leads at 65.5%; the closest result above is Gemini 3.8 Flash (high) at 19.6%. It covers hard multi-hour software engineering. The board median is 29.5%, so GLM-5.3-Flash (max) is 11.4 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GLM-5.3-Flash (max) lands below the middle on Mystery Games: 8.0%, #47 of 59, with 78% of the field scoring higher. GPT-6 Astra (max) leads at 84.0%; the closest result above is Qwen3-235B-A22B-Thinking-2507 at 9.0%. It covers deductive reasoning puzzles. The board median is 23.0%, so GLM-5.3-Flash (max) is 15.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GLM-5.3-Flash (max) lands below the middle on CursorBench: 36.8%, #12 of 15, with 73% of the field scoring higher. Claude Opus 5.5 (max) leads at 57.8%; the closest result above is Gemini 3.8 Flash (high) at 39.6%. It covers real-world coding-agent tasks. The board median is 41.6%, so GLM-5.3-Flash (max) is 4.8 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GDP.pdf is a weak spot. GLM-5.3-Flash (max) scores 14.0%, #21 of 25, behind 80% of the models measured. The leader, GPT-6 Astra (max), reaches 34.2%. It covers economically valuable document tasks. The board median is 22.8%, so GLM-5.3-Flash (max) is 8.8 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

The full table above also lists the other 4 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: weak, ahead of 30% of the field on average across 4 benchmarks. Its best board here is FrontierCode (#8 of 15) and its weakest is CursorBench (#12 of 15). Reasoning, maths and science: solid, ahead of 62% of the field on average across 6 benchmarks. Its best board here is GPQA Diamond (#35 of 275) and its weakest is FrontierMath T4 (#36 of 50). General knowledge and chat: uneven, ahead of 45% of the field on average across 3 benchmarks. Its best board here is Surface Evolver (#5 of 13) and its weakest is Mystery Games (#47 of 59).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so GLM-5.3-Flash (max) costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen3.7 Flash ($0.030, 50.5). Similar scores cost more on Qwen3.8 Flash (xhigh) ($0.113) and GLM-4.5-Air ($0.125).

Release history

The Z.ai (Zhipu AI) model released before it was GLM-5.3 (max) on 2026-08-14, which scores 53.8.

Reading these numbers

The 14 scores on this page come from 2 sources: Epoch AI Benchmarking Hub (13), ARC Prize official leaderboard (1). 1 of them come from an official leaderboard and 13 from datasets or third-party runs. The leaderboard also lists 4 other reasoning-effort settings of this model, which redirect here. This page shows the highest.

GLM-5.3-Flash (max) is also listed at 4 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

• #245 GLM-4.5-FP8 (49.6)
• #246 GLM-5.1 (49.6)
• #247 Opus 4.5 (Thinking, 64K) (49.6)
• #249 grok-4.20-0309-reasoning (49.6)
• #250 Grok-2 (49.6)
• #251 ollama/qwen2.5:32b-instruct-q8_0 (49.6)

More from Z.ai (Zhipu AI)

• GLM-5.3 (max) (#28, 53.8)
• GLM-5 (#118, 51.5)
• GLM-5.1 (#246, 49.6)
• GLM-4.7 (Novita) (#355, 48.4)
• GLM-4.7 (#410, 47.7)

Compare it on the live leaderboard → · How the aggregate score works