Skip to main content
Radar

Provider changes are easy to miss. Radar watches releases, pricing, deprecations, and incidents at the source.Provider changes are easy to miss.

See Radar
BenchLM
Advertise

Model comparison

GPT-5.2 vs GPT-5.3 Codex

Updated August 7, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.

GPT-5.2

OpenAI

57.8/100

Estimated · Public rank #74

90% interval 49.5–66.0

GPT-5.3 Codex

OpenAI

65.8/100

Supported · Public rank #30

90% interval 62.4–69.1

GPT-5.3 Codex has the higher public score estimate, 65.75 versus 57.77, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.

6 results are shared. Category rows based on different benchmark sets are marked directional and do not name a winner.

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.

No workload recommendation clears the current evidence threshold.

Use the matched evidence, workload costs, and sourced specifications below instead of treating a point score as a universal answer.

Show secondary and unsupported calls

  • Coding work

    Code generation, repair, and software-engineering tasks

    Not enough matched evidence

    The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.

    Confidence: limited

  • Agentic work

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

    Not enough matched evidence

    The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.

    Confidence: limited

  • Long documents

    Prompts that approach the documented context limit

    No clear pick

    The documented context windows are equal.

    Confidence: documented

  • Chat turn cost

    1K fresh input + 500 output tokens

    No clear pick

    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

    No clear pick

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

    Confidence: rate-fallback

  • Repository review cost

    50K fresh input + 3K output tokens

    No clear pick

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

    Confidence: listed-rates

What is actually comparable

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

Shared results
6
GPT-5.2 only
9
GPT-5.3 Codex only
2
Like-for-like categories
0 / 8

2 categories use different evidence sets. Those rows remain visible for coverage context but do not name a winner.

Category results, on a stated basis

Each row states whether both averages use the same weighted benchmark set. Directional and not-comparable rows remain visible, but they never receive a winner in this template.

Agentic

Directional only
GPT-5.2
55.7
GPT-5.3 Codex
71.4
Weighted basis
2 vs 2 rows
Reading
Directional only

Coding

Directional only
GPT-5.2
70.6
GPT-5.3 Codex
67.2
Weighted basis
2 vs 3 rows
Reading
Directional only

Reasoning

Not comparable
GPT-5.2
52.9
GPT-5.3 Codex
Not measured
Weighted basis
1 vs 0 rows
Reading
Not comparable

Knowledge

Not comparable
GPT-5.2
92.4
GPT-5.3 Codex
Not measured
Weighted basis
1 vs 0 rows
Reading
Not comparable

Math

Not comparable
GPT-5.2
35.2
GPT-5.3 Codex
Not measured
Weighted basis
2 vs 0 rows
Reading
Not comparable

Multilingual

Not comparable
GPT-5.2
Not measured
GPT-5.3 Codex
Not measured
Weighted basis
0 vs 0 rows
Reading
Not comparable

Multimodal

Not comparable
GPT-5.2
80.4
GPT-5.3 Codex
Not measured
Weighted basis
2 vs 0 rows
Reading
Not comparable

Instruction following

Not comparable
GPT-5.2
Not measured
GPT-5.3 Codex
Not measured
Weighted basis
0 vs 0 rows
Reading
Not comparable

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.

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-5.2
$0.00875
Fits in one request
GPT-5.3 Codex
$0.00875
Fits in one request

Modeled costs are equal

Costs use the listed standard API rates.

Repository review

50K fresh input + 3K output tokens

GPT-5.2
$0.1295
Fits in one request
GPT-5.3 Codex
$0.1295
Fits in one request

Modeled costs are equal

Costs use the listed standard API rates.

Cache-heavy agent loop

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

GPT-5.2
$0.525
Fits in one request
Cached input priced at the published list-input rate
GPT-5.3 Codex
$0.525
Fits in one request
Cached input priced at the published list-input rate

Modeled costs are equal

GPT-5.2 has no published cached-input rate, so cached tokens use its listed input rate. GPT-5.3 Codex has no published cached-input rate, so cached tokens use its listed input rate.

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-5.2

Not published

GPT-5.3 Codex

Not published

Provider availability

GPT-5.2

Not sourced

GPT-5.3 Codex

Generally Available · OpenAI Responses API

OpenAI model catalog

Reasoning profile

GPT-5.2

Reasoning

GPT-5.3 Codex

Reasoning

Weight access

GPT-5.2

Proprietary

GPT-5.3 Codex

Proprietary

License

GPT-5.2

Proprietary

GPT-5.3 Codex

Proprietary

Release date

GPT-5.2

2025-12-11

GPT-5.3 Codex

2026-02-05

If you already use one of these models
Deployment change
Both entries list OpenAI as the provider. Confirm endpoint, model ID, limits, and feature support before switching.
Quality signal
GPT-5.3 Codex has the higher public score estimate, 65.75 versus 57.77, but the 90% score intervals overlap.
Workload cost
Repository review: $0.1295 vs $0.1295. Cache-heavy agent loop: $0.525 vs $0.525.
Context tradeoff
Both models list 400K.

Run the same representative tasks against both endpoints before changing production traffic.

Benchmark evidence

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

Browse raw public benchmark evidence17 rows

Agentic

  • BrowseComp

    GPT-5.265.8%
    Source
    GPT-5.3 Codex—

    Not directly comparable

  • OSWorld-Verified

    GPT-5.247.3%
    Source
    GPT-5.3 Codex64.7%
    Source

    GPT-5.3 Codex leads this result

  • Gert Labs

    Shared source
    GPT-5.246.54%
    GPT-5.3 Codex57.47%

    GPT-5.3 Codex leads this result

  • JobBench

    Shared source
    GPT-5.234.3%
    GPT-5.3 Codex33.7%

    GPT-5.2 leads this result

  • Terminal-Bench 2.0

    GPT-5.2—
    GPT-5.3 Codex77.3%
    Source

    Not directly comparable

Coding

  • SWE-bench Verified

    GPT-5.280%
    Source
    GPT-5.3 Codex85%
    Source

    GPT-5.3 Codex leads this result

  • SWE-bench Pro

    GPT-5.255.6%
    Source
    GPT-5.3 Codex56.8%
    Source

    GPT-5.3 Codex leads this result

  • Vibe Code Bench

    Shared source
    GPT-5.253.50%
    GPT-5.3 Codex61.77%

    GPT-5.3 Codex leads this result

  • SWE-Rebench

    GPT-5.2—
    GPT-5.3 Codex58.2%
    Source

    Not directly comparable

Reasoning

  • ARC-AGI-2

    GPT-5.252.9%
    Source
    GPT-5.3 Codex—

    Not directly comparable

Knowledge

  • GPQA

    GPT-5.292.4%
    Source
    GPT-5.3 Codex—

    Not directly comparable

Math

  • FrontierMath v2 (Tiers 1-3)

    GPT-5.240.700%
    Source
    GPT-5.3 Codex—

    Not directly comparable

  • FrontierMath v2 (Tier 4)

    GPT-5.218.800%
    Source
    GPT-5.3 Codex—

    Not directly comparable

Multimodal

  • MMMU-Pro

    GPT-5.279.5%
    Source
    GPT-5.3 Codex—

    Not directly comparable

  • MathVision

    GPT-5.283.0%
    Source
    GPT-5.3 Codex—

    Not directly comparable

  • CharXiv

    GPT-5.282.1%
    Source
    GPT-5.3 Codex—

    Not directly comparable

  • V*

    GPT-5.275.9%
    Source
    GPT-5.3 Codex—

    Not directly comparable

Frequently asked questions

Which is better, GPT-5.2 or GPT-5.3 Codex?

GPT-5.3 Codex has the higher public score estimate, 65.75 versus 57.77, 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-5.2 or GPT-5.3 Codex?

The current coding averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.

Which is better for agentic tasks, GPT-5.2 or GPT-5.3 Codex?

The current agentic tasks averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.

Which costs less, GPT-5.2 or GPT-5.3 Codex?

For the stated presets, chat costs $0.00875 on GPT-5.2 and $0.00875 on GPT-5.3 Codex; repository review costs $0.1295 and $0.1295; the cache-heavy agent loop costs $0.525 and $0.525. GPT-5.2 has no published cached-input rate, so cached tokens use its listed input rate. GPT-5.3 Codex has no published cached-input rate, so cached tokens use its listed input rate.

Which has the larger context window, GPT-5.2 or GPT-5.3 Codex?

Both models list the same context window, 400K.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated August 7, 2026

Watch GPT-5.2 vs GPT-5.3 Codex

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

Read a sample issue

Join 2,000+ readers.