Skip to main content

Model comparison

GPT-5.3 Codex vs Qwen3.5 397B

Data verified

Head-to-head evidence from 16 shared benchmark results across 6 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.

66.69/100
Margin
9.7pts
← winning
57.01/100
2 category wins0 category wins

Public leaderboard positions: GPT-5.3 Codex #26 (Supported); Qwen3.5 397B #71 (Estimated). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.

Evidence parity. GPT-5.3 Codex and Qwen3.5 397B share 16 comparable benchmark results. 2 of 8 categories are comparable. 5 results are unique to GPT-5.3 Codex; 39 to Qwen3.5 397B.

Updated July 20, 2026
Shared results
16
GPT-5.3 Codex only
5
Qwen3.5 397B only
39
Comparable categories
2 / 8

Pick GPT-5.3 Codex if you want the stronger benchmark profile. Qwen3.5 397B only becomes the better choice if you want the cheaper token bill or you would rather avoid the extra latency and token burn of a reasoning model.

Confidence note. This is a partial-evidence comparison with 16 shared benchmark results across 6 evidence categories; 2 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.

Why this result

GPT-5.3 Codex is clearly ahead on the BenchAlign aggregate, 66.69 to 57.01. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

GPT-5.3 Codex's sharpest advantage is in agentic, where it averages 71.4 against 56.5. The single biggest benchmark swing on the page is Terminal-Bench 2.0, 77.3% to 52.5%.

GPT-5.3 Codex is also the more expensive model on tokens at $1.75 input / $14.00 output per 1M tokens, versus $0.60 input / $3.60 output per 1M tokens for Qwen3.5 397B. That is roughly 3.9x on output cost alone. GPT-5.3 Codex is the reasoning model in the pair, while Qwen3.5 397B is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. GPT-5.3 Codex gives you the larger context window at 400K, compared with 128K for Qwen3.5 397B.

Category breakdown

Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.

Category scores and score margins for GPT-5.3 Codex and Qwen3.5 397B
CategoryGPT-5.3 CodexΔQwen3.5 397B
AgenticGPT-5.3 Codex71.4Margin 14.9Qwen3.5 397B56.5
CodingGPT-5.3 Codex67.2Margin 0.7Qwen3.5 397B66.5
ReasoningGPT-5.3 CodexNot measuredMarginNo overlapQwen3.5 397B63.2
KnowledgeGPT-5.3 CodexNot measuredMarginNo overlapQwen3.5 397B56.6
MathGPT-5.3 CodexNot measuredMarginNo overlapQwen3.5 397B90.6
MultilingualGPT-5.3 CodexNot measuredMarginNo overlapQwen3.5 397B84.7
MultimodalGPT-5.3 CodexNot measuredMarginNo overlapQwen3.5 397B79.6
Inst. FollowingGPT-5.3 CodexNot measuredMarginNo overlapQwen3.5 397B92.6

Decisive benchmark drivers

The largest measured benchmark gaps in this matchup, with exact reported values.

More
A · GPT-5.3 CodexB · Qwen3.5 397B
  1. Terminal-Bench 2.0

    Agentic
    Source ↗
    A 77.3%B 52.5%
    Winner: GPT-5.3 CodexΔ 24.8
    Terminal-Bench 2.0: GPT-5.3 Codex scored 77.3%; Qwen3.5 397B scored 52.5%. GPT-5.3 Codex wins this benchmark.
  2. SWE-bench Verified

    Coding
    Source ↗
    A 85%B 76.2%
    Winner: GPT-5.3 CodexΔ 8.8
    SWE-bench Verified: GPT-5.3 Codex scored 85%; Qwen3.5 397B scored 76.2%. GPT-5.3 Codex wins this benchmark.
  3. SWE-bench Pro

    Coding
    Source ↗
    A 56.8%B 50.9%
    Winner: GPT-5.3 CodexΔ 5.9
    SWE-bench Pro: GPT-5.3 Codex scored 56.8%; Qwen3.5 397B scored 50.9%. GPT-5.3 Codex wins this benchmark.

Operational comparison

Runtime and commercial metrics are compared only when both models have a complete sourced value.

MetricGPT-5.3 CodexQwen3.5 397BComparison
Input / output priceUSD per 1M tokensGPT-5.3 Codex$1.75 input / $14 outputQwen3.5 397B$0.6 input / $3.6 outputQwen3.5 397B has the lower combined listed price.
Generation speedtokens per secondGPT-5.3 Codex79 tok/sQwen3.5 397B96 tok/sQwen3.5 397B has the higher measured throughput.
First-answer latencyseconds to first tokenGPT-5.3 Codex88.26 sQwen3.5 397B2.44 sQwen3.5 397B reaches the first token sooner.
Context windowmaximum listed tokensGPT-5.3 Codex400KQwen3.5 397B128KGPT-5.3 Codex lists the larger context window.

Benchmark Deep Dive

AgenticGPT-5.3 Codex wins
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
Terminal-Bench 2.0Source 77.3%52.5%GPT-5.3 Codex leads
OSWorld-VerifiedSource 64.7%Not comparable
τ²-bench resultsSource 86%95.6%Qwen3.5 397B leads
Gert LabsSource 57.47%46.76%GPT-5.3 Codex leads
JobBenchSource 33.7%Not comparable
BrowseCompSource 62%Not comparable
Claw-EvalSource 56.8%Not comparable
QwenClawBenchSource 51.8%Not comparable
τ³-bench resultsSource 68.4%Not comparable
VITA-BenchSource 43.7%Not comparable
DeepPlanningSource 37.6%Not comparable
ToolathlonSource 36.3%Not comparable
MCP AtlasSource 46.1%Not comparable
MCP-TasksSource 74.2%Not comparable
WideResearchSource 74.0%Not comparable
ResearchClawBenchSource 14.2%Not comparable
AA Agentic IndexSource 19.9%Not comparable
APEX-Agents-AASource 15.3%Not comparable
GDPval-AASource 23.1%Not comparable
GDPval-AASource 962Not comparable
CodingGPT-5.3 Codex wins
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
SWE-bench VerifiedSource 85%76.2%GPT-5.3 Codex leads
SWE-bench ProSource 56.8%50.9%GPT-5.3 Codex leads
SWE-RebenchSource 58.2%Not comparable
Vibe Code BenchSource 61.77%Not comparable
AA-SciCodeSource 53.2%42.0%GPT-5.3 Codex leads
LiveCodeBench v6Source 83.6%Not comparable
AA Coding IndexSource 48.2%Not comparable
Reasoning
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
AA-LCRSource 74.0%65.7%GPT-5.3 Codex leads
CritPtSource 16.9%1.7%GPT-5.3 Codex leads
LongBench v2Source 63.2%Not comparable
AI-NeedleSource 68.7%Not comparable
Knowledge
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
Artificial Analysis Intelligence IndexSource 44.3%33.7%GPT-5.3 Codex leads
AA-GPQA DiamondSource 91.5%89.3%GPT-5.3 Codex leads
AA-HLESource 39.9%27.3%GPT-5.3 Codex leads
AA-Omniscience IndexSource 9.9%-29.8%GPT-5.3 Codex leads
AA-Omniscience AccuracySource 51.8%31.4%GPT-5.3 Codex leads
AA-Omniscience Hallucination RateSource 86.9%89.1%GPT-5.3 Codex leads
GPQASource 88.4%Not comparable
SuperGPQASource 70.4%Not comparable
MMLU-ProSource 87.8%Not comparable
MMLU-ReduxSource 94.9%Not comparable
C-EvalSource 93%Not comparable
HLESource 28.7%Not comparable
Math
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
AIME26Source 93.3%Not comparable
HMMT Feb 2025Source 94.8%Not comparable
HMMT Nov 2025Source 92.7%Not comparable
HMMT Feb 2026Source 87.9%Not comparable
MMAnswerBenchSource 80.9%Not comparable
Multilingual
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
MMLU-ProXSource 84.7%Not comparable
NOVA-63Source 59.1%Not comparable
Multimodal
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
AA-MMMU-ProSource 78.5%77.3%GPT-5.3 Codex leads
Design Arena WebsiteSource 1195Not comparable
MMMU-ProSource 79%Not comparable
MathVisionSource 88.6%Not comparable
CharXivSource 80.8%Not comparable
VideoMMMUSource 84.7%Not comparable
ScreenSpot ProSource 65.6%Not comparable
V*Source 95.8%Not comparable
Inst. Following
BenchmarkGPT-5.3 CodexQwen3.5 397BResult
AA-IFBenchSource 75.4%78.8%Qwen3.5 397B leads
IFEvalSource 92.6%Not comparable
Frequently Asked Questions (3)

Which is better, GPT-5.3 Codex or Qwen3.5 397B?

GPT-5.3 Codex is ahead on BenchLM's BenchAlign leaderboard, 66.69 to 57.01. The biggest single separator in this matchup is Terminal-Bench 2.0, where the scores are 77.3% and 52.5%.

Which is better for coding, GPT-5.3 Codex or Qwen3.5 397B?

GPT-5.3 Codex has the edge for coding in this comparison, averaging 67.2 versus 66.5. Inside this category, AA-SciCode is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, GPT-5.3 Codex or Qwen3.5 397B?

GPT-5.3 Codex has the edge for agentic tasks in this comparison, averaging 71.4 versus 56.5. Inside this category, Terminal-Bench 2.0 is the benchmark that creates the most daylight between them.

Related Comparisons

Last updated: July 20, 2026

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.