GPT-5.6 Luna vs Step 3.7 Flash

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.

Decision reading

GPT-5.6 Luna has the higher public score, 65.6 versus 41.41, and the 90% score intervals do not overlap. 5 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.

Model A

GPT-5.6 Luna

OpenAI

65.6/100

Supported · Public rank #25

90% interval 60.5–70.7

Model B

Step 3.7 Flash

StepFun

41.41/100

Estimated · Public rank #109

90% interval 29.9–52.9

Shared results: 5
GPT-5.6 Luna only: 24
Step 3.7 Flash only: 6
Like-for-like categories: 0 / 8

Supported: GPT-5.6 Luna · Estimated: Step 3.7 FlashHow 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: documented
Chat turn cost
1K fresh input + 500 output tokens
Step 3.7 Flash
Step 3.7 Flash 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-5.6 Luna
GPT-5.6 Luna has the lower estimated token cost for this stated workload. Step 3.7 Flash 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
Step 3.7 Flash
Step 3.7 Flash 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
Step 3.7 Flash is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
Step 3.7 Flash 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.

Pick the job you would use them for

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

64.5GPT-5.6 Luna32.4Step 3.7 Flash

Directional only · BenchAlign v5.7

GPT-5.6 Luna scores higher, but at least one score rests on Estimated evidence or a different benchmark set. Directional only, no winner.

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

Open the coding leaderboard Shortlist models for this job

Same basis rules as the category table below

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.

BrowseCompAgentic
Normalized gap 7.5
GPT-5.6 Luna:83.3%
Step 3.7 Flash:75.8%
GPT-5.6 Luna source Step 3.7 Flash source
SWE-bench ProCoding
Normalized gap 6.4
GPT-5.6 Luna:62.7%
Step 3.7 Flash:56.3%
GPT-5.6 Luna source Step 3.7 Flash source

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

Directional only

GPT-5.6 Luna: 55.3; Supported · #28/105
Step 3.7 Flash: 35.2; Estimated · #56/105
Basis: BenchAlign v5.7 lane · 9 vs 7 public rows
Reading: Directional only

Coding

Directional only

GPT-5.6 Luna: 64.5; Supported · #9/135
Step 3.7 Flash: 32.4; Estimated · #81/135
Basis: BenchAlign v5.7 lane · 7 vs 2 public rows
Reading: Directional only

Knowledge

Directional only

GPT-5.6 Luna: 64.6; Supported · #22/158
Step 3.7 Flash: 43.1; Estimated · #78/158
Basis: BenchAlign v5.7 lane · 6 vs 0 public rows
Reading: Directional only

Reasoning

Not comparable

GPT-5.6 Luna: 54.7; #18/19
Step 3.7 Flash: 72.7; Unranked · 2 rankable rows
Basis: Provisional lane · 2 vs 0 weighted rows
Reading: Not comparable

Multimodal

Not comparable

GPT-5.6 Luna: 67.1; #22/50
Step 3.7 Flash: 71.0; Unranked · 3 rankable rows
Basis: Provisional lane · 1 vs 0 weighted rows
Reading: Not comparable

Multilingual

Not comparable

GPT-5.6 Luna: Not ranked
Step 3.7 Flash: Not ranked
Basis: Provisional lane · 0 vs 0 weighted rows
Reading: Not comparable

Instruction following

Not comparable

GPT-5.6 Luna: Not ranked
Step 3.7 Flash: 80.6; #53/124
Basis: Provisional lane · 0 vs 0 weighted rows
Reading: Not comparable

Math

Not comparable

GPT-5.6 Luna: 94.2; Unranked · 3 rankable rows
Step 3.7 Flash: Not ranked
Basis: Provisional lane · 2 vs 0 weighted rows
Reading: Not comparable

Category scores with the server-provided evidence basis for GPT-5.6 Luna and Step 3.7 Flash
Category	GPT-5.6 Luna	Step 3.7 Flash	Basis	Reading
Agentic	55.3Supported · #28/105	35.2Estimated · #56/105	Directional onlyBenchAlign v5.7 lane · 9 vs 7 public rows	Directional only
Coding	64.5Supported · #9/135	32.4Estimated · #81/135	Directional onlyBenchAlign v5.7 lane · 7 vs 2 public rows	Directional only
Knowledge	64.6Supported · #22/158	43.1Estimated · #78/158	Directional onlyBenchAlign v5.7 lane · 6 vs 0 public rows	Directional only
Reasoning	54.7#18/19	72.7Unranked · 2 rankable rows	Not comparableProvisional lane · 2 vs 0 weighted rows	Not comparable
Multimodal	67.1#22/50	71.0Unranked · 3 rankable rows	Not comparableProvisional lane · 1 vs 0 weighted rows	Not comparable
Multilingual	Not ranked	Not ranked	Not comparableProvisional lane · 0 vs 0 weighted rows	Not comparable
Instruction following	Not ranked	80.6#53/124	Not comparableProvisional lane · 0 vs 0 weighted rows	Not comparable
Math	94.2Unranked · 3 rankable rows	Not ranked	Not comparableProvisional lane · 2 vs 0 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-5.6 Luna: $0.0008; Fits in one request
Step 3.7 Flash: $0.00077; Fits in one request

Step 3.7 Flash has the lower modeled cost

Costs use the listed standard API rates.

Repository review

50K fresh input + 3K output tokens

GPT-5.6 Luna: $0.0136; Fits in one request
Step 3.7 Flash: $0.01345; Fits in one request

Step 3.7 Flash 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-5.6 Luna: $0.02; Fits in one request
Step 3.7 Flash: $0.0555; Fits in one request; Cached input priced at the published list-input rate

GPT-5.6 Luna has the lower modeled cost

Step 3.7 Flash 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-5.6 LunaStep 3.7 Flash

Context window

Maximum documented context; output-token limits may be lower.

GPT-5.6 Luna

1.05M

OpenAI model catalog

Step 3.7 Flash

256K

API model ID

GPT-5.6 Luna

gpt-5.6-luna

OpenAI model catalog

Step 3.7 Flash

Not sourced

Cached-input rate

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

GPT-5.6 Luna

$0.02 per 1M cached input tokens

OpenAI pricing

Step 3.7 Flash

Not published

Documented inputs

GPT-5.6 Luna

text, image

OpenAI model catalog

Step 3.7 Flash

Not sourced

Documented outputs

GPT-5.6 Luna

text

OpenAI model catalog

Step 3.7 Flash

Not sourced

Provider availability

GPT-5.6 Luna

Generally Available · OpenAI Responses API

OpenAI model catalog

Step 3.7 Flash

Not sourced

Reasoning profile

GPT-5.6 Luna

Reasoning

Step 3.7 Flash

Reasoning

Weight access

GPT-5.6 Luna

Proprietary

Step 3.7 Flash

Open Weight

License

GPT-5.6 Luna

Proprietary

Step 3.7 Flash

Open Weight

Release date

GPT-5.6 Luna

2026-07-09

Step 3.7 Flash

2026-05-29

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-5.6 Luna has the higher public score, 65.6 versus 41.41, and the 90% score intervals do not overlap.
Workload cost: Repository review: $0.0136 vs $0.01345. Cache-heavy agent loop: $0.02 vs $0.0555.
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-5.6 Luna or Step 3.7 Flash?

GPT-5.6 Luna has the higher public score, 65.6 versus 41.41, 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-5.6 Luna or Step 3.7 Flash?

GPT-5.6 Luna scores higher for coding on the public lane, 64.5 to 32.4. Step 3.7 Flash is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like. BenchLM does not name a winner for coding from a directional reading; read the shared benchmark rows directly and test the models on the same task set.

Which is better for agentic tasks, GPT-5.6 Luna or Step 3.7 Flash?

GPT-5.6 Luna scores higher for agentic tasks on the public lane, 55.3 to 35.2. Step 3.7 Flash 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-5.6 Luna or Step 3.7 Flash?

For the stated presets, chat costs $0.0008 on GPT-5.6 Luna and $0.00077 on Step 3.7 Flash; repository review costs $0.0136 and $0.01345; the cache-heavy agent loop costs $0.02 and $0.0555. Step 3.7 Flash has no published cached-input rate, so cached tokens use its listed input rate.

Which has the larger context window, GPT-5.6 Luna or Step 3.7 Flash?

GPT-5.6 Luna has the larger documented context window: 1.05M, compared with 256K.

Benchmark evidence

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

Browse raw public benchmark evidence35 rows

Agentic

Terminal-Bench 3.0
GPT-5.6 Luna14.3%
Source
Step 3.7 Flash—
Not directly comparable
Terminal-Bench 2.1
GPT-5.6 Luna84.7%
Source
Step 3.7 Flash59.5%
Source
GPT-5.6 Luna leads this result
BrowseComp
GPT-5.6 Luna83.3%
Source
Step 3.7 Flash75.8%
Source
GPT-5.6 Luna leads this result
OSWorld 2.0
GPT-5.6 Luna45.6%
Source
Step 3.7 Flash—
Not directly comparable
CyberGym
GPT-5.6 Luna77.9%
Source
Step 3.7 Flash—
Not directly comparable
ExploitGym
GPT-5.6 Luna12.4%
Source
Step 3.7 Flash—
Not directly comparable
Toolathlon
GPT-5.6 Luna53.4%
Source
Step 3.7 Flash49.5%
Source
GPT-5.6 Luna leads this result
Terminal-Bench 2.1 (Vals)
GPT-5.6 Luna79.0%
Source
Step 3.7 Flash—
Not directly comparable
ApprenticeBench
GPT-5.6 Luna7%
Source
Step 3.7 Flash—
Not directly comparable
DeepSearchQA
GPT-5.6 Luna—
Step 3.7 Flash92.8%
Source
Not directly comparable
Claw-Eval
GPT-5.6 Luna—
Step 3.7 Flash67.1%
Source
Not directly comparable
HLE w/ tools
GPT-5.6 Luna—
Step 3.7 Flash47.2%
Source
Not directly comparable
Gert Labs
GPT-5.6 Luna—
Step 3.7 Flash51.57%
Source
Not directly comparable

Coding

SWE-bench Pro
GPT-5.6 Luna62.7%
Source
Step 3.7 Flash56.3%
Source
GPT-5.6 Luna leads this result
Terminal-Bench 2.1
GPT-5.6 Luna84.7%
Source
Step 3.7 Flash59.5%
Source
GPT-5.6 Luna leads this result
DeepSWE
GPT-5.6 Luna67.2%
Source
Step 3.7 Flash—
Not directly comparable
FrontierCode 1.1 Extended
GPT-5.6 Luna55.1%
Source
Step 3.7 Flash—
Not directly comparable
cursorBench32
GPT-5.6 Luna61.1%
Source
Step 3.7 Flash—
Not directly comparable
VulcanBench v3
GPT-5.6 Luna85.5%
Source
Step 3.7 Flash—
Not directly comparable
SWE-bench (Vals)
GPT-5.6 Luna93.0%
Source
Step 3.7 Flash—
Not directly comparable

Reasoning

ARC-AGI-2
GPT-5.6 Luna59.5%
Source
Step 3.7 Flash—
Not directly comparable
ARC-AGI-3
GPT-5.6 Luna0.2%
Source
Step 3.7 Flash—
Not directly comparable

Multimodal

MMMU-Pro
GPT-5.6 Luna78.4%
Source
Step 3.7 Flash—
Not directly comparable
MMMU-Pro w/ Python
GPT-5.6 Luna79.5%
Source
Step 3.7 Flash—
Not directly comparable
SimpleVQA
GPT-5.6 Luna—
Step 3.7 Flash79.2%
Source
Not directly comparable
V*
GPT-5.6 Luna—
Step 3.7 Flash95.3%
Source
Not directly comparable

Knowledge

GPQA
GPT-5.6 Luna92.3%
Source
Step 3.7 Flash—
Not directly comparable
GPQA-D
GPT-5.6 Luna92.3%
Source
Step 3.7 Flash—
Not directly comparable
HealthBench Professional
GPT-5.6 Luna55.7%
Source
Step 3.7 Flash—
Not directly comparable
HealthBench Hard
GPT-5.6 Luna32.0%
Source
Step 3.7 Flash—
Not directly comparable
GPQA Diamond (Vals)
GPT-5.6 Luna91.7%
Source
Step 3.7 Flash—
Not directly comparable
MMLU-Pro (Vals)
GPT-5.6 Luna86.0%
Source
Step 3.7 Flash—
Not directly comparable

Math

FrontierMath (legacy)
GPT-5.6 Luna78.6%
Source
Step 3.7 Flash—
Not directly comparable
FrontierMath v2 (Tiers 1-3)
GPT-5.6 Luna78.600%
Source
Step 3.7 Flash—
Not directly comparable
FrontierMath v2 (Tier 4)
GPT-5.6 Luna58.500%
Source
Step 3.7 Flash—
Not directly comparable

35 public results · 5 shared

Related comparisons

Compare API pricing Read the methodology Open the model selector

Watch GPT-5.6 Luna vs Step 3.7 Flash

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 24, 2026

GPT-5.6 Luna vs Step 3.7 Flash

Decision reading

Which one for your work

Long documents

Chat turn cost

Cache-heavy agent loop cost

Repository review cost

Coding work

Agentic work

Which one for a specific job

What is actually comparable

Shape of the matched evidence

Category results, on a stated basis

Agentic

Coding

Knowledge

Reasoning

Multimodal

Multilingual

Instruction following

Math

What each workload costs

Chat turn

Repository review

Cache-heavy agent loop

Specification differences

Context window

API model ID

Cached-input rate

Documented inputs

Documented outputs

Provider availability

Reasoning profile

Weight access

License

Release date

If you already use one of these models

Questions

Which is better, GPT-5.6 Luna or Step 3.7 Flash?

Which is better for coding, GPT-5.6 Luna or Step 3.7 Flash?

Which is better for agentic tasks, GPT-5.6 Luna or Step 3.7 Flash?

Which costs less, GPT-5.6 Luna or Step 3.7 Flash?

Which has the larger context window, GPT-5.6 Luna or Step 3.7 Flash?

Benchmark evidence

Agentic

Coding

Reasoning

Multimodal

Knowledge

Math

Related comparisons

Watch GPT-5.6 Luna vs Step 3.7 Flash