Model comparison

GPT-5.4 nano vs Qwen3.8 27B

Qwen3.8 27B is the stronger model overall, scoring 46.0 to 41.9 on the Noometry Index. GPT-5.4 nano costs 2.4× less per token, which makes it the better buy when Qwen3.8 27B's lead doesn't matter for your workload.

Last verified October 10, 2026. 26 shared benchmarks.

GPT-5.4 nano OpenAI

41.9

Rank #125 Confirmed

Qwen3.8 27B

Alibaba (Qwen)

46.0

Rank #68 Confirmed

Summary

They share 26 benchmarks with published results for both. GPT-5.4 nano scores higher in 2 categories and Qwen3.8 27B in 7 categories; 8 gaps are clear of the uncertainty.
The widest gap is in reasoning, where Qwen3.8 27B leads 41.0 to 23.7.
The biggest single-benchmark swing is ARC-AGI-2: 5.7% for GPT-5.4 nano and 42.4% for Qwen3.8 27B.
GPT-5.4 nano is cheaper at $0.20 / $1.25 per million input/output tokens, against $0.99 / $1.49 for Qwen3.8 27B.
GPT-5.4 nano accepts more context: 400K tokens versus 262K.
Qwen3.8 27B has downloadable open weights; the other is API-only.

Side by side

GPT-5.4 nano and Qwen3.8 27B specifications
	GPT-5.4 nano	Qwen3.8 27B
Provider	OpenAI	Alibaba (Qwen)
Noometry Index	41.9	46.0
Released	2026-03-17	2026-08-14
Weights	Proprietary	Open
Context window	400K	262K
Max output	128K	33K
Input $ / M tokens	$0.20	$0.99
Output $ / M tokens	$1.25	$1.49
Results tracked	40	31

Sponsored placements are available on pages like this one. Advertise on Noometry

Category by category

Coding Qwen3.8 27B leads

GPT-5.4 nano: 43.6 (#84), Qwen3.8 27B: 50.5 (#44)

Coding benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
SciCode	46.9%	46.6%
LMArena Coding	1405	1482
LMArena WebDev	—	1593
WeirdML	49.2%	—
ALE-Bench	1,005	—

Agentic & Tool Use Not comparable

GPT-5.4 nano: —, Qwen3.8 27B: 32.9 (#57)

Agentic & Tool Use benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
APEX-Agents	—	47.5%

Reasoning Qwen3.8 27B leads

GPT-5.4 nano: 23.7 (#173), Qwen3.8 27B: 41.0 (#54)

Reasoning benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
ARC-AGI-2	5.7%	42.4%
ARC-AGI-1	51.5%	87.5%
CritPt	9.3%	5.4%
LMArena Hard Prompts	1381	1460
DTBench	80.3%	88%
LMCA	36.9%	41.4%
Epoch Capabilities Index	145.81	149.38
Kagi LLM Benchmark	39.7%	—
NYT Connections (extended)	—	54.5%
Chess Puzzles	30%	—
Mystery Game Puzzles	9%	—
Surface Evolver Bench	—	45%
ForecastBench	57.3	—

Math GPT-5.4 nano leads

GPT-5.4 nano: 40.9 (#88), Qwen3.8 27B: 37.1 (#161)

Math benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
ProofBench	5%	16%
LMArena Math	1406	1456
FrontierMath (Tiers 1-3)	44.9%	—
FrontierMath Tier 4	12.2%	—
OTIS Mock AIME 2024-2025	87.8%	—
FrontierMath (Feb 2025 set)	25.9%	—
FrontierMath Tier 4 (v1)	6.3%	—

Knowledge Too close to call

GPT-5.4 nano: 41.9 (#103), Qwen3.8 27B: 41.6 (#109)

Knowledge benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
LMArena Expert	1396	1482
GPQA Diamond	78.5%	—
SimpleQA Verified	11.7%	—
Vectara Hallucination Rate	3.1%	—

Multimodal Qwen3.8 27B leads

GPT-5.4 nano: 36.7 (#78), Qwen3.8 27B: 41.3 (#37)

Multimodal benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
LMArena Vision	1196	1271

Multilingual Qwen3.8 27B leads

GPT-5.4 nano: 48.6 (#140), Qwen3.8 27B: 53.7 (#60)

Multilingual benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
LMArena Non-English	1359	1430
LMArena Chinese	1392	1504
LMArena French	1396	1465
LMArena German	1367	1438
LMArena Japanese	1343	1384
LMArena Korean	1320	1393
LMArena Russian	1363	1415
LMArena Spanish	1371	1448

Instruction Following Qwen3.8 27B leads

GPT-5.4 nano: 71.9 (#144), Qwen3.8 27B: 75.8 (#53)

Instruction Following benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
LMArena Instruction Following	1362	1439

Long Context Qwen3.8 27B leads

GPT-5.4 nano: 41.6 (#137), Qwen3.8 27B: 44.3 (#70)

Long Context benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
LMArena Longer Query	1366	1450

Writing & Preference Qwen3.8 27B leads

GPT-5.4 nano: 55.7 (#142), Qwen3.8 27B: 65.8 (#43)

Writing & Preference benchmarks
Benchmark	GPT-5.4 nano	Qwen3.8 27B
LMArena Text	1372	1441
LMArena Creative Writing	1314	1384
LMArena Multi-Turn	1382	1441
EQ-Bench Creative Writing	—	1671

Frequently asked questions

Is GPT-5.4 nano better than Qwen3.8 27B?

Which is cheaper, GPT-5.4 nano or Qwen3.8 27B?

GPT-5.4 nano is cheaper. It lists at $0.20 per million input tokens and $1.25 per million output tokens; Qwen3.8 27B lists at $0.99 and $1.49.

Is GPT-5.4 nano or Qwen3.8 27B better for coding?

Qwen3.8 27B scores higher on coding benchmarks: 50.5 versus 43.6 in the Noometry coding category.

Which has the bigger context window?

GPT-5.4 nano does, with 400K tokens against 262K.

How many benchmarks do GPT-5.4 nano and Qwen3.8 27B share?

26 benchmarks have published results for both models. GPT-5.4 nano has 40 scored results on Noometry and Qwen3.8 27B has 31.