Coding work
Code generation, repair, and software-engineering tasks
GPT-5.6 Luna
GPT-5.6 Luna leads on the public coding lane, 64.5 to 47.1, with Supported evidence for both models, although the 90% intervals overlap.
Confidence: limited
GPT-5.2-Codex vs GPT-5.6 Luna
Updated September 24, 2026. Rank says GPT-5.6 Luna is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.
Both of these models will change. Get the price, version and retirement notices for the pair, each with its source. Follow model changes
Share or export
Decision reading
GPT-5.6 Luna has the higher public score, 65.6 versus 50.15, and the 90% score intervals do not overlap. 1 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
- Shared results
- 1
- GPT-5.2-Codex only
- 4
- GPT-5.6 Luna only
- 28
- Like-for-like categories
- 1 / 8
Estimated: GPT-5.2-Codex · Supported: GPT-5.6 LunaHow 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
GPT-5.6 Luna
GPT-5.6 Luna has the larger documented context window.
Confidence: documentedChat turn cost
1K fresh input + 500 output tokens
GPT-5.6 Luna
GPT-5.6 Luna has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Show secondary and unsupported calls
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
GPT-5.6 Luna
GPT-5.6 Luna has the lower estimated token cost for this stated workload. GPT-5.2-Codex has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallbackRepository review cost
50K fresh input + 3K output tokens
GPT-5.6 Luna
GPT-5.6 Luna has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-ratesAgentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
GPT-5.2-Codex is scored on Estimated evidence for agentic, so the reading is directional rather than like-for-like.
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.
47.1GPT-5.2-Codex64.5GPT-5.6 Luna
Like-for-like · BenchAlign v5.7
GPT-5.6 Luna leads the like-for-like coding row, although the 90% intervals overlap.
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.
4 GPT-5.2-Codex only
1 Shared
28 GPT-5.6 Luna only
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.
A shared-evidence shape is not available.
BenchLM does not draw a radar or infer missing axes when the matched evidence is too sparse.
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.
Coding
Like-for-like- GPT-5.2-Codex
- 47.1
- Supported · #43/135
- GPT-5.6 Luna
- 64.5
- Supported · #9/135
- Basis
- BenchAlign v5.7 lane · 3 vs 7 public rows
- Reading
- GPT-5.6 Luna leads · intervals overlap
Agentic
Directional only- GPT-5.2-Codex
- 34.2
- Estimated · #57/105
- GPT-5.6 Luna
- 55.3
- Supported · #28/105
- Basis
- BenchAlign v5.7 lane · 2 vs 9 public rows
- Reading
- Directional only
Knowledge
Directional only- GPT-5.2-Codex
- 52.0
- Estimated · #54/158
- GPT-5.6 Luna
- 64.6
- Supported · #22/158
- Basis
- BenchAlign v5.7 lane · 0 vs 6 public rows
- Reading
- Directional only
Reasoning
Not comparable- GPT-5.2-Codex
- 78.6
- Unranked · 2 rankable rows
- GPT-5.6 Luna
- 54.7
- #18/19
- Basis
- Provisional lane · 0 vs 2 weighted rows
- Reading
- Not comparable
Multimodal
Not comparable- GPT-5.2-Codex
- 72.5
- Unranked · 1 rankable row
- GPT-5.6 Luna
- 67.1
- #22/50
- Basis
- Provisional lane · 0 vs 1 weighted rows
- Reading
- Not comparable
Multilingual
Not comparable- GPT-5.2-Codex
- Not ranked
- GPT-5.6 Luna
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Instruction following
Not comparable- GPT-5.2-Codex
- 92.4
- #3/124
- GPT-5.6 Luna
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Math
Not comparable- GPT-5.2-Codex
- Not ranked
- GPT-5.6 Luna
- 94.2
- Unranked · 3 rankable rows
- Basis
- Provisional lane · 0 vs 2 weighted rows
- Reading
- Not comparable
| Category | GPT-5.2-Codex | GPT-5.6 Luna | Basis | Reading |
|---|---|---|---|---|
| Coding | 47.1Supported · #43/135 | 64.5Supported · #9/135 | Like-for-likeBenchAlign v5.7 lane · 3 vs 7 public rows | GPT-5.6 Luna leads · intervals overlap |
| Agentic | 34.2Estimated · #57/105 | 55.3Supported · #28/105 | Directional onlyBenchAlign v5.7 lane · 2 vs 9 public rows | Directional only |
| Knowledge | 52.0Estimated · #54/158 | 64.6Supported · #22/158 | Directional onlyBenchAlign v5.7 lane · 0 vs 6 public rows | Directional only |
| Reasoning | 78.6Unranked · 2 rankable rows | 54.7#18/19 | Not comparableProvisional lane · 0 vs 2 weighted rows | Not comparable |
| Multimodal | 72.5Unranked · 1 rankable row | 67.1#22/50 | Not comparableProvisional lane · 0 vs 1 weighted rows | Not comparable |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Instruction following | 92.4#3/124 | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Math | Not ranked | 94.2Unranked · 3 rankable rows | Not comparableProvisional lane · 0 vs 2 weighted rows | 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-5.2-Codex
- $0.00875
- Fits in one request
- GPT-5.6 Luna
- $0.0008
- Fits in one request
GPT-5.6 Luna has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- GPT-5.2-Codex
- $0.1295
- Fits in one request
- GPT-5.6 Luna
- $0.0136
- Fits in one request
GPT-5.6 Luna 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-5.2-Codex
- $0.525
- Fits in one request
- Cached input priced at the published list-input rate
- GPT-5.6 Luna
- $0.02
- Fits in one request
GPT-5.6 Luna has the lower modeled cost
GPT-5.2-Codex 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.
SpecificationGPT-5.2-CodexGPT-5.6 Luna
Context window
Maximum documented context; output-token limits may be lower.
GPT-5.2-Codex
400K
GPT-5.6 Luna
1.05M
OpenAI model catalogAPI model ID
GPT-5.2-Codex
Not sourced
GPT-5.6 Luna
gpt-5.6-luna
OpenAI model catalogCached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
GPT-5.2-Codex
Not published
GPT-5.6 Luna
$0.02 per 1M cached input tokens
OpenAI pricingDocumented inputs
GPT-5.2-Codex
Not sourced
GPT-5.6 Luna
text, image
OpenAI model catalogDocumented outputs
GPT-5.2-Codex
Not sourced
GPT-5.6 Luna
Provider availability
GPT-5.2-Codex
Not sourced
GPT-5.6 Luna
Generally Available · OpenAI Responses API
OpenAI model catalogReasoning profile
GPT-5.2-Codex
Reasoning
GPT-5.6 Luna
Reasoning
Weight access
GPT-5.2-Codex
Proprietary
GPT-5.6 Luna
Proprietary
License
GPT-5.2-Codex
Proprietary
GPT-5.6 Luna
Proprietary
Release date
GPT-5.2-Codex
2025-12-18
GPT-5.6 Luna
2026-07-09
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-5.6 Luna has the higher public score, 65.6 versus 50.15, and the 90% score intervals do not overlap.
- Workload cost
- Repository review: $0.1295 vs $0.0136. Cache-heavy agent loop: $0.525 vs $0.02.
- Context tradeoff
- GPT-5.6 Luna has the larger documented window (1.05M).
Run the same representative tasks against both endpoints before changing production traffic.
Questions
Which is better, GPT-5.2-Codex or GPT-5.6 Luna?
GPT-5.6 Luna has the higher public score, 65.6 versus 50.15, and the 90% score intervals do not overlap. That is the clearest overall quality signal in the current public data.
Which is better for coding, GPT-5.2-Codex or GPT-5.6 Luna?
GPT-5.6 Luna leads the public coding lane, 64.5 to 47.1, with Supported evidence for both models, although the 90% intervals overlap.
Which is better for agentic tasks, GPT-5.2-Codex or GPT-5.6 Luna?
GPT-5.6 Luna scores higher for agentic tasks on the public lane, 55.3 to 34.2. GPT-5.2-Codex is scored on Estimated evidence for agentic tasks, so the reading is directional rather than like-for-like. BenchLM does not name a winner for agentic tasks from a directional reading; read the shared benchmark rows directly and test the models on the same task set.
Which costs less, GPT-5.2-Codex or GPT-5.6 Luna?
For the stated presets, chat costs $0.00875 on GPT-5.2-Codex and $0.0008 on GPT-5.6 Luna; repository review costs $0.1295 and $0.0136; the cache-heavy agent loop costs $0.525 and $0.02. GPT-5.2-Codex has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, GPT-5.2-Codex or GPT-5.6 Luna?
GPT-5.6 Luna has the larger documented context window: 1.05M, 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 evidence33 rows
Agentic
Gert Labs
GPT-5.2-Codex51.79%SourceGPT-5.6 Luna—Not directly comparable
JobBench
GPT-5.2-Codex26.0%SourceGPT-5.6 Luna—Not directly comparable
Terminal-Bench 3.0
GPT-5.2-Codex—GPT-5.6 Luna14.3%SourceNot directly comparable
Terminal-Bench 2.1
GPT-5.2-Codex—GPT-5.6 Luna84.7%SourceNot directly comparable
BrowseComp
GPT-5.2-Codex—GPT-5.6 Luna83.3%SourceNot directly comparable
OSWorld 2.0
GPT-5.2-Codex—GPT-5.6 Luna45.6%SourceNot directly comparable
CyberGym
GPT-5.2-Codex—GPT-5.6 Luna77.9%SourceNot directly comparable
ExploitGym
GPT-5.2-Codex—GPT-5.6 Luna12.4%SourceNot directly comparable
Toolathlon
GPT-5.2-Codex—GPT-5.6 Luna53.4%SourceNot directly comparable
Terminal-Bench 2.1 (Vals)
GPT-5.2-Codex—GPT-5.6 Luna79.0%SourceNot directly comparable
ApprenticeBench
GPT-5.2-Codex—GPT-5.6 Luna7%SourceNot directly comparable
Coding
Vibe Code Bench
GPT-5.2-Codex37.91%SourceGPT-5.6 Luna—Not directly comparable
LiveCodeBench (Vals)
GPT-5.2-Codex88.0%SourceGPT-5.6 Luna—Not directly comparable
SWE-bench (Vals)
GPT-5.2-Codex72.4%SourceGPT-5.6 Luna93.0%SourceGPT-5.6 Luna leads this result
SWE-bench Pro
GPT-5.2-Codex—GPT-5.6 Luna62.7%SourceNot directly comparable
Terminal-Bench 2.1
GPT-5.2-Codex—GPT-5.6 Luna84.7%SourceNot directly comparable
DeepSWE
GPT-5.2-Codex—GPT-5.6 Luna67.2%SourceNot directly comparable
FrontierCode 1.1 Extended
GPT-5.2-Codex—GPT-5.6 Luna55.1%SourceNot directly comparable
cursorBench32
GPT-5.2-Codex—GPT-5.6 Luna61.1%SourceNot directly comparable
VulcanBench v3
GPT-5.2-Codex—GPT-5.6 Luna85.5%SourceNot directly comparable
Reasoning
Multimodal
Knowledge
GPQA
GPT-5.2-Codex—GPT-5.6 Luna92.3%SourceNot directly comparable
GPQA-D
GPT-5.2-Codex—GPT-5.6 Luna92.3%SourceNot directly comparable
HealthBench Professional
GPT-5.2-Codex—GPT-5.6 Luna55.7%SourceNot directly comparable
HealthBench Hard
GPT-5.2-Codex—GPT-5.6 Luna32.0%SourceNot directly comparable
GPQA Diamond (Vals)
GPT-5.2-Codex—GPT-5.6 Luna91.7%SourceNot directly comparable
MMLU-Pro (Vals)
GPT-5.2-Codex—GPT-5.6 Luna86.0%SourceNot directly comparable
Math
33 public results · 1 shared
Watch GPT-5.2-Codex vs GPT-5.6 Luna
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.
Last updated September 24, 2026