Skip to main content
BenchLM

GPT-5.4 mini vs Qwen3.5 Plus

Updated September 28, 2026. Rank says GPT-5.4 mini is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.

Share or export
Share on XLinkedInSocial cardCSVJSON

Decision reading

GPT-5.4 mini has the higher public score estimate, 54.95 versus 48.92, but the 90% score intervals overlap. Treat that as a lead, not a settled winner. 3 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.

Model A
OpenAI logo

OpenAI

54.95/100

Supported · Public rank #62

90% interval 44.8–65.1

Model B
Alibaba logo

Alibaba

48.92/100

Estimated · Public rank #86

90% interval 37.4–60.4

Shared results
3
GPT-5.4 mini only
18
Qwen3.5 Plus only
1
Like-for-like categories
0 / 8
Supported: GPT-5.4 mini · Estimated: Qwen3.5 PlusHow the comparison works

Which one for your work

Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.

  • Long documents

    Prompts that approach the documented context limit

    Qwen3.5 Plus

    Qwen3.5 Plus has the larger documented context window.

    Confidence: documented
  • Chat turn cost

    1K fresh input + 500 output tokens

    Qwen3.5 Plus

    Qwen3.5 Plus has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.

    Confidence: listed-rates
  • Cache-heavy agent loop cost

    200K cached + 20K fresh input + 10K output tokens

    GPT-5.4 mini

    GPT-5.4 mini has the lower estimated token cost for this stated workload. Qwen3.5 Plus has no published cached-input rate, so cached tokens use its listed input rate.

    Confidence: rate-fallback
Show secondary and unsupported calls
  • Repository review cost

    50K fresh input + 3K output tokens

    Qwen3.5 Plus

    Qwen3.5 Plus has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.

    Confidence: listed-rates
  • Coding work

    Code generation, repair, and software-engineering tasks

    Not enough matched evidence

    Qwen3.5 Plus is not ranked on the public lane for coding, so no winner is named for coding.

    Confidence: limited
  • Agentic work

    Tool use, computer use, and multi-step task completion

    Not enough matched evidence

    Qwen3.5 Plus is not ranked on the public lane for agentic, so no winner is named for agentic.

    Confidence: limited

Which one for a specific job

Choose a job from the LLM Selector's task catalog to see the category row it rests on, under the same basis rules as the table below. A directional row stays directional; choosing a job never creates a winner.

The same task catalog as the LLM Selector. Each job names the evidence surface it rests on; nothing here adds a new score.

36.9GPT-5.4 mini—Qwen3.5 Plus

Not comparable · BenchAlign v5.7

The coding row is not comparable on the public lane: at least one model is not measured or not ranked there.

Coding scores combine specific tasks and setups. Match the editor, harness, and effort to your workflow.

Same basis rules as the category table below

What is actually comparable

Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.

Shape of the matched evidence

Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.

Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.

  • FrontierMath v2 (Tiers 1-3)Math

    Normalized gap 7.2
    GPT-5.4 mini:28.280%
    Qwen3.5 Plus:21.034%
  • FrontierMath v2 (Tier 4)Math

    Normalized gap 0.0
    GPT-5.4 mini:2.080%
    Qwen3.5 Plus:2.083%
Bars run 0–100 on each benchmark’s normalized display scale

Category results, on a stated basis

Each row shows the public-lane category score for both models: the BenchAlign v5.7 lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.

Agentic

Not comparable
GPT-5.4 mini
34.0
Supported · #71/117
Qwen3.5 Plus
Not ranked
Basis
BenchAlign v5.7 lane · 6 vs 1 public rows
Reading
Not comparable

Coding

Not comparable
GPT-5.4 mini
36.9
Supported · #71/142
Qwen3.5 Plus
Not ranked
Basis
BenchAlign v5.7 lane · 4 vs 1 public rows
Reading
Not comparable

Reasoning

Not comparable
GPT-5.4 mini
40.0
Unranked · 4 rankable rows
Qwen3.5 Plus
Not ranked
Basis
Provisional lane · 1 vs 0 weighted rows
Reading
Not comparable

Multimodal

Not comparable
GPT-5.4 mini
58.3
#33/50
Qwen3.5 Plus
Not ranked
Basis
Provisional lane · 1 vs 0 weighted rows
Reading
Not comparable

Knowledge

Not comparable
GPT-5.4 mini
48.4
Supported · #65/168
Qwen3.5 Plus
Not ranked
Basis
BenchAlign v5.7 lane · 5 vs 0 public rows
Reading
Not comparable

Multilingual

Not comparable
GPT-5.4 mini
Not ranked
Qwen3.5 Plus
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Instruction following

Not comparable
GPT-5.4 mini
88.5
#24/124
Qwen3.5 Plus
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Math

Not comparable
GPT-5.4 mini
44.3
Unranked · 2 rankable rows
Qwen3.5 Plus
39.3
Unranked · 2 rankable rows
Basis
Provisional lane · 2 vs 2 weighted rows
Reading
Not comparable

Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign v5.7) differ from the provisional-lane categories. Unranked scores sit on the provisional lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.

Supported evidence per lane · bars run 0–100Methodology

What each workload costs

Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.

Chat turn

1K fresh input + 500 output tokens

GPT-5.4 mini
$0.003
Fits in one request
Qwen3.5 Plus
$0.0016
Fits in one request

Qwen3.5 Plus has the lower modeled cost

Costs use the listed standard API rates.

Repository review

50K fresh input + 3K output tokens

GPT-5.4 mini
$0.051
Fits in one request
Qwen3.5 Plus
$0.0272
Fits in one request

Qwen3.5 Plus has the lower modeled cost

Costs use the listed standard API rates.

Cache-heavy agent loop

200K cached + 20K fresh input + 10K output tokens

GPT-5.4 mini
$0.075
Fits in one request
Qwen3.5 Plus
$0.112
Fits in one request
Cached input priced at the published list-input rate

GPT-5.4 mini has the lower modeled cost

Qwen3.5 Plus has no published cached-input rate, so cached tokens use its listed input rate.

Cached input falls back to the list input rate only where a cached rate is unpublished

Specification differences

Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.

Cached-input rate

A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.

GPT-5.4 mini

$0.075 per 1M cached input tokens

OpenAI pricing

Qwen3.5 Plus

Not published

Provider availability

GPT-5.4 mini

Generally Available · OpenAI Responses API

OpenAI model catalog

Qwen3.5 Plus

Not sourced

Reasoning profile

GPT-5.4 mini

Reasoning

Qwen3.5 Plus

Reasoning

Weight access

GPT-5.4 mini

Proprietary

Qwen3.5 Plus

Proprietary

License

GPT-5.4 mini

Proprietary

Qwen3.5 Plus

Proprietary

Release date

GPT-5.4 mini

2026-03-17

Qwen3.5 Plus

2026-03-04

If you already use one of these models

Deployment change
The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
Quality signal
GPT-5.4 mini has the higher public score estimate, 54.95 versus 48.92, but the 90% score intervals overlap.
Workload cost
Repository review: $0.051 vs $0.0272. Cache-heavy agent loop: $0.075 vs $0.112.
Context tradeoff
Qwen3.5 Plus has the larger documented window (1M).
Run the same representative tasks against both endpoints before changing production traffic.

Questions

Which is better, GPT-5.4 mini or Qwen3.5 Plus?

GPT-5.4 mini has the higher public score estimate, 54.95 versus 48.92, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.

Which is better for coding, GPT-5.4 mini or Qwen3.5 Plus?

Qwen3.5 Plus is not ranked on the public lane for coding, so no winner is named for coding.

Which is better for agentic tasks, GPT-5.4 mini or Qwen3.5 Plus?

Qwen3.5 Plus is not ranked on the public lane for agentic tasks, so no winner is named for agentic tasks.

Which costs less, GPT-5.4 mini or Qwen3.5 Plus?

For the stated presets, chat costs $0.003 on GPT-5.4 mini and $0.0016 on Qwen3.5 Plus; repository review costs $0.051 and $0.0272; the cache-heavy agent loop costs $0.075 and $0.112. Qwen3.5 Plus has no published cached-input rate, so cached tokens use its listed input rate.

Which has the larger context window, GPT-5.4 mini or Qwen3.5 Plus?

Qwen3.5 Plus has the larger documented context window: 1M, compared with 400K.

Benchmark evidence

The full public result ledger is available for audit without forcing a wide desktop table onto a phone.

Browse raw public benchmark evidence22 rows

Agentic

  • Terminal-Bench 2.0

    GPT-5.4 mini60%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • OSWorld-Verified

    GPT-5.4 mini72.1%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • MCP Atlas

    GPT-5.4 mini57.7%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • Toolathlon

    GPT-5.4 mini42.9%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • τ²-bench results

    GPT-5.4 mini93.4%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • Terminal-Bench 2.1 (Vals)

    GPT-5.4 mini54.7%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • JobBench

    GPT-5.4 mini—
    Qwen3.5 Plus18.5%
    Source

    Not directly comparable

Coding

  • Vibe Code Bench

    Shared source
    GPT-5.4 mini47.97%
    Qwen3.5 Plus15.74%

    GPT-5.4 mini leads this result

  • FrontierCode 1.1 Main

    GPT-5.4 mini27.0%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • LiveCodeBench (Vals)

    GPT-5.4 mini81.5%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • SWE-bench (Vals)

    GPT-5.4 mini73.0%
    Source
    Qwen3.5 Plus—

    Not directly comparable

Reasoning

  • ARC-AGI-1

    GPT-5.4 mini63.70%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • ARC-AGI-2

    GPT-5.4 mini18.9%
    Source
    Qwen3.5 Plus—

    Not directly comparable

Multimodal

  • MMMU-Pro

    GPT-5.4 mini76.6%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • MMMU-Pro w/ Python

    GPT-5.4 mini78%
    Source
    Qwen3.5 Plus—

    Not directly comparable

Knowledge

  • GPQA

    GPT-5.4 mini88%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • HLE

    GPT-5.4 mini41.5%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • HLE w/o tools

    GPT-5.4 mini28.2%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • GPQA Diamond (Vals)

    GPT-5.4 mini83.1%
    Source
    Qwen3.5 Plus—

    Not directly comparable

  • MMLU-Pro (Vals)

    GPT-5.4 mini84.6%
    Source
    Qwen3.5 Plus—

    Not directly comparable

Math

  • FrontierMath v2 (Tiers 1-3)

    Shared source
    GPT-5.4 mini28.280%
    Qwen3.5 Plus21.034%

    GPT-5.4 mini leads this result

  • FrontierMath v2 (Tier 4)

    Shared source
    GPT-5.4 mini2.080%
    Qwen3.5 Plus2.083%

    Qwen3.5 Plus leads this result

22 public results · 3 shared

Watch GPT-5.4 mini vs Qwen3.5 Plus

One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.

Read a sample issue

Join 2,000+ readers.

Last updated September 28, 2026