Skip to main content
BenchLM
Data

GPT-5.4 vs Qwen3.8-27B

Updated November 19, 2026. Rank says GPT-5.4 is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.

Share or export
Share on XLinkedInAPI/MCP

Decision reading

GPT-5.4 has the higher public score estimate, 62.7 versus 55, but the 90% score intervals overlap. Treat that as a lead, not a settled winner. 15 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

62.7/100

Supported · Public rank #46

90% interval 58.6–66.8

Model B
Alibaba logo

Alibaba

55/100

Supported · Public rank #72

90% interval 49.5–60.5

API price per 1M tokens

GPT-5.4: $2.50 input / $15 output

Qwen3.8-27B: No comparable API price

Context window

GPT-5.4: 1.05M

Qwen3.8-27B: 262K

Released

GPT-5.4: March 5, 2026

Qwen3.8-27B: August 5, 2026

Shared results
15
GPT-5.4 only
36
Qwen3.8-27B only
23
Like-for-like categories
5 / 8
Supported: GPT-5.4 and Qwen3.8-27BHow 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.

  • Coding work

    Code generation, repair, and software-engineering tasks

    GPT-5.4

    GPT-5.4 has the higher public coding point estimate, 60.6 to 55.4, with Supported evidence for both models. The conditional ranges do not establish rank confidence.

    Confidence: limited
  • Agentic work

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

    GPT-5.4

    GPT-5.4 has the higher public agentic point estimate, 61.6 to 58.2, with Supported evidence for both models. The conditional ranges do not establish rank confidence.

    Confidence: limited
  • Long documents

    Prompts that approach the documented context limit

    GPT-5.4

    GPT-5.4 has the larger documented context window.

    Confidence: documented
Show secondary and unsupported calls
  • Chat turn cost

    1K fresh input + 500 output tokens

    Not enough matched evidence

    A complete comparable API-rate estimate is not available for both models.

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

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

    Not enough matched evidence

    A complete comparable API-rate estimate is not available for both models.

    Confidence: listed-rates
  • Repository review cost

    50K fresh input + 3K output tokens

    Not enough matched evidence

    A complete comparable API-rate estimate is not available for both models.

    Confidence: listed-rates

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.

60.6GPT-5.455.4Qwen3.8-27B

Like-for-like · BenchAlign v6.0

GPT-5.4 has the higher coding point estimate. Conditional score ranges do not establish rank confidence.

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

Same basis rules as the category table below

Or describe your own job

The LLM Selector reads what you need a model for and returns a shortlist, ranked in the same order the job pages publish.

Describe the task and constraints. This selector recommends models; it cannot carry out the task or access your files and accounts.

We keep what you type here, drafts included, for 180 days, and after that only the job and requirements read from it. Nothing is tied to your IP address or account. Leave out names and other personal details. Privacy

What is actually comparable

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

2 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.

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.

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 v6.0 lane in every category, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence. Directional and not-comparable rows remain visible, but they never receive a winner.

Agentic

Like-for-like
GPT-5.4
61.6
Supported · #52/177
Qwen3.8-27B
58.2
Supported · #64/177
Basis
BenchAlign v6.0 lane · 16 vs 9 public rows
Reading
GPT-5.4 leads · intervals overlap

Coding

Like-for-like
GPT-5.4
60.6
Supported · #51/177
Qwen3.8-27B
55.4
Supported · #65/177
Basis
BenchAlign v6.0 lane · 10 vs 9 public rows
Reading
GPT-5.4 leads · intervals overlap

Reasoning

Like-for-like
GPT-5.4
65.2
Supported · #30/160
Qwen3.8-27B
50.2
Supported · #74/160
Basis
BenchAlign v6.0 lane · 4 vs 0 public rows
Reading
GPT-5.4 leads · intervals overlap

Multimodal

Like-for-like
GPT-5.4
58.6
Supported · #50/115
Qwen3.8-27B
63.4
Supported · #38/115
Basis
BenchAlign v6.0 lane · 10 vs 13 public rows
Reading
Qwen3.8-27B leads · intervals overlap

Instruction following

Like-for-like
GPT-5.4
60.1
Supported · #34/142
Qwen3.8-27B
52.0
Supported · #60/142
Basis
BenchAlign v6.0 lane · 0 vs 1 public rows
Reading
GPT-5.4 leads · intervals overlap

Knowledge

Directional only
GPT-5.4
59.5
Supported · #46/156
Qwen3.8-27B
52.9
Estimated · #71/156
Basis
BenchAlign v6.0 lane · 9 vs 6 public rows
Reading
Directional only

Math

Directional only
GPT-5.4
63.1
Supported · #25/124
Qwen3.8-27B
51.6
Estimated · #59/124
Basis
BenchAlign v6.0 lane · 2 vs 0 public rows
Reading
Directional only

Multilingual

Not comparable
GPT-5.4
Not ranked
Qwen3.8-27B
Not ranked
Basis
BenchAlign v6.0 lane · 0 vs 0 public rows
Reading
Not comparable

Ranks count the models scored in each category’s BenchAlign v6.0 lane, so denominators differ by category. 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
$0.01
Fits in one request
Qwen3.8-27B
Self-hosted; infrastructure cost varies
Fits in one request

Qwen3.8-27B has no comparable published API token rate.

Repository review

50K fresh input + 3K output tokens

GPT-5.4
$0.17
Fits in one request
Qwen3.8-27B
Self-hosted; infrastructure cost varies
Fits in one request

Qwen3.8-27B has no comparable published API token rate.

Cache-heavy agent loop

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

GPT-5.4
$0.25
Fits in one request
Qwen3.8-27B
Self-hosted; infrastructure cost varies
Fits in one request
Cached-input rate unavailable

Qwen3.8-27B has no comparable published API token rate.

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

Price against score, among all priced models

Overall score against output price per 1M tokens, one model per family. Pricing as of November 18, 2026.

GPT-5.4

20406080100$0.10$0.50$1$5$10$50$100Output price, $ per 1M tokens (log)Overall score
  • 14 models score at least as high as GPT-5.4 for the same price or less. The nearest is Kimi K3 (score 65.5, also $15.00 per 1M output tokens).
  • Qwen3.8-27B has no public API price, so it is not plotted.

Gray dots are the other priced models. The dark step line is the efficiency frontier: for a model under it, another model costs no more and scores at least as high.

See GPT-5.4 on the full price vs performance chart

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

$0.25 per 1M cached input tokens

OpenAI pricing

Qwen3.8-27B

No comparable hosted API rate

Qwen3.8-27B model card

Provider availability

GPT-5.4

Not sourced

Qwen3.8-27B

Not sourced

Reasoning profile

GPT-5.4

Reasoning

Qwen3.8-27B

Reasoning

Weight access

GPT-5.4

Proprietary

Qwen3.8-27B

Open Weight

License

GPT-5.4

Proprietary

Qwen3.8-27B

Open Weight

Release date

GPT-5.4

2026-03-05

Qwen3.8-27B

2026-08-05

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 has the higher public score estimate, 62.7 versus 55, but the 90% score intervals overlap.
Workload cost
A complete comparable API-rate estimate is not available for both models.
Context tradeoff
GPT-5.4 has the larger documented window (1.05M).
Run the same representative tasks against both endpoints before changing production traffic.

Questions

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

GPT-5.4 has the higher public score estimate, 62.7 versus 55, 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 or Qwen3.8-27B?

GPT-5.4 has the higher public coding point estimate, 60.6 to 55.4, with Supported evidence for both models. The conditional ranges do not establish rank confidence.

Which is better for agentic tasks, GPT-5.4 or Qwen3.8-27B?

GPT-5.4 has the higher public agentic tasks point estimate, 61.6 to 58.2, with Supported evidence for both models. The conditional ranges do not establish rank confidence.

Which costs less, GPT-5.4 or Qwen3.8-27B?

Both models do not have comparable published API token rates, so this page does not name a universal price winner.

Which has the larger context window, GPT-5.4 or Qwen3.8-27B?

GPT-5.4 has the larger documented context window: 1.05M, compared with 262K.

Self-host vs API cost

Estimates at 50,000 req/day · 1000 tokens/req average.

GPT-5.4
API / mo$13,125
Self-host / moNot listed
Break-even—
Proprietary model — self-hosting not applicable.
Qwen3.8-27B
API / mo (DeepInfra)$1,519
Self-host / mo$298
Break-even10M/day
Model the full break-even

Benchmark evidence

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

Browse raw public benchmark evidence74 rows

Agentic

  • LifeVal

    GPT-5.473.0%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • Terminal-Bench 2.0

    GPT-5.475.1%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • CyberGym

    GPT-5.479.0%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • BrowseComp

    GPT-5.482.7%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • OSWorld-Verified

    GPT-5.475%
    Source
    Qwen3.8-27B84.3%
    Source

    Qwen3.8-27B leads this result

  • MCP Atlas

    GPT-5.470.6%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • Toolathlon

    GPT-5.454.6%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • τ²-bench results

    GPT-5.498.9%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • Claw-Eval

    GPT-5.460.3%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • DeepSearchQA

    GPT-5.473.6%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • Gert Labs

    GPT-5.464.89%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ResearchClawBench

    GPT-5.415.3%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • JobBench

    GPT-5.438.9%
    Source
    Qwen3.8-27B33.4%
    Source

    GPT-5.4 leads this result

  • ExploitGym

    GPT-5.46.0%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ApprenticeBench

    GPT-5.411%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • MazeBench

    GPT-5.40%
    Source
    Qwen3.8-27B0%
    Source

    Tie

  • Terminal-Bench 2.1

    GPT-5.4—
    Qwen3.8-27B73.0%
    Source

    Not directly comparable

  • CoWorkBench

    GPT-5.4—
    Qwen3.8-27B70.7%
    Source

    Not directly comparable

  • Agents' Last Exam

    GPT-5.4—
    Qwen3.8-27B42.9%
    Source

    Not directly comparable

  • WebArena-Verified

    GPT-5.4—
    Qwen3.8-27B64.8%
    Source

    Not directly comparable

  • AndroidWorld

    GPT-5.4—
    Qwen3.8-27B81.9%
    Source

    Not directly comparable

  • Terminal-Bench 2.1 (Vals)

    GPT-5.4—
    Qwen3.8-27B58.4%
    Source

    Not directly comparable

Coding

  • LiveCodeBench Pro

    GPT-5.487.5%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • SWE-bench Pro

    GPT-5.457.7%
    Source
    Qwen3.8-27B61.7%
    Source

    Qwen3.8-27B leads this result

  • React Native Evals

    GPT-5.485.3%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • Terminal-Bench 2.0

    GPT-5.475.1%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • Vibe Code Bench

    GPT-5.467.42%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • PostTrainBench v1.1

    GPT-5.419.0%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • LiveCodeBench (Vals)

    GPT-5.484.1%
    Source
    Qwen3.8-27B84.0%
    Source

    GPT-5.4 leads this result

  • SWE-bench (Vals)

    GPT-5.478.2%
    Source
    Qwen3.8-27B86.0%
    Source

    Qwen3.8-27B leads this result

  • PostTrainBench

    GPT-5.417.8%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • PostTrainBench v1

    GPT-5.420.2%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • Terminal-Bench 2.1

    GPT-5.4—
    Qwen3.8-27B73.0%
    Source

    Not directly comparable

  • NL2Repo

    GPT-5.4—
    Qwen3.8-27B42.3%
    Source

    Not directly comparable

  • DeepSWE

    GPT-5.4—
    Qwen3.8-27B42.2%
    Source

    Not directly comparable

  • LiveCodeBench v6

    GPT-5.4—
    Qwen3.8-27B90.3%
    Source

    Not directly comparable

  • VulcanBench v3

    GPT-5.4—
    Qwen3.8-27B82.6%
    Source

    Not directly comparable

  • Bug Hunt Bench

    GPT-5.4—
    Qwen3.8-27B15.0 fixes
    Source

    Not directly comparable

Reasoning

  • BioBasicsBench

    GPT-5.433.2%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ARC-AGI-2

    GPT-5.474.0%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ARC-AGI-3

    GPT-5.40.2%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ARC-AGI-1

    GPT-5.493.67%
    Source
    Qwen3.8-27B—

    Not directly comparable

Multimodal

  • ParseBench-1

    Shared source
    GPT-5.461.72%
    Qwen3.8-27B69.03%

    Qwen3.8-27B leads this result

  • MMMU-Pro

    GPT-5.481.2%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • OfficeQA Pro

    GPT-5.453.2%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • MMMU-Pro w/ Python

    GPT-5.482.1%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • CharXiv

    GPT-5.482.8%
    Source
    Qwen3.8-27B90.2%
    Source

    Qwen3.8-27B leads this result

  • ERQA

    GPT-5.465.4%
    Source
    Qwen3.8-27B65.5%
    Source

    Qwen3.8-27B leads this result

  • SimpleVQA

    GPT-5.461.1%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ScreenSpot Pro

    GPT-5.485.4%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ZeroBench

    GPT-5.441.0%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • MedXpertQA (MM)

    GPT-5.477.1%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • ParseBench-2

    GPT-5.4—
    Qwen3.8-27B74.40%
    Source

    Not directly comparable

  • ExtractBench (LlamaIndex)

    GPT-5.4—
    Qwen3.8-27B89.75%
    Source

    Not directly comparable

  • MathVision

    GPT-5.4—
    Qwen3.8-27B90.0%
    Source

    Not directly comparable

  • MathVision w/ Python

    GPT-5.4—
    Qwen3.8-27B94.6%
    Source

    Not directly comparable

  • BabyVision

    GPT-5.4—
    Qwen3.8-27B65.7%
    Source

    Not directly comparable

  • BabyVision w/ Python

    GPT-5.4—
    Qwen3.8-27B85.6%
    Source

    Not directly comparable

  • Vision2Web

    GPT-5.4—
    Qwen3.8-27B62.9%
    Source

    Not directly comparable

  • CharXiv w/o tools

    GPT-5.4—
    Qwen3.8-27B83.7%
    Source

    Not directly comparable

  • OmniDocBench 1.5

    GPT-5.4—
    Qwen3.8-27B91.1%
    Source

    Not directly comparable

  • RealWorldQA

    GPT-5.4—
    Qwen3.8-27B85.9%
    Source

    Not directly comparable

Knowledge

  • GPQA

    GPT-5.492.8%
    Source
    Qwen3.8-27B89.2%
    Source

    GPT-5.4 leads this result

  • HLE

    GPT-5.452.1%
    Source
    Qwen3.8-27B30.8%
    Source

    GPT-5.4 leads this result

  • HLE w/o tools

    GPT-5.439.8%
    Source
    Qwen3.8-27B30.8%
    Source

    GPT-5.4 leads this result

  • GPQA-D

    GPT-5.492.8%
    Source
    Qwen3.8-27B89.2%
    Source

    GPT-5.4 leads this result

  • HealthBench Hard

    GPT-5.440.1%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • MedXpertQA (Text)

    GPT-5.459.6%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • HealthBench Professional

    GPT-5.448.1%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • GPQA Diamond (Vals)

    GPT-5.491.7%
    Source
    Qwen3.8-27B88.9%
    Source

    GPT-5.4 leads this result

  • MMLU-Pro (Vals)

    GPT-5.487.5%
    Source
    Qwen3.8-27B84.3%
    Source

    GPT-5.4 leads this result

Instruction following

  • IFBench

    GPT-5.4—
    Qwen3.8-27B79.5%
    Source

    Not directly comparable

Math

  • FrontierMath v2 (Tiers 1-3)

    GPT-5.478.596%
    Source
    Qwen3.8-27B—

    Not directly comparable

  • FrontierMath v2 (Tier 4)

    GPT-5.449.000%
    Source
    Qwen3.8-27B—

    Not directly comparable

74 public results · 15 shared

Watch GPT-5.4 vs Qwen3.8-27B

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

Read a sample issue

Join 7,000+ subscribers.

Last updated November 19, 2026