Skip to main content
Radar

Every change to the models you run, with its source and its date. Releases, price changes, retirements, API changes, and incidents.Every change to the models you run, with its source.

Follow model changes
BenchLM

GPT-6 Luna vs Grok 4.20

Decision reading

Grok 4.20 has the higher public score estimate, 67.97 versus 62.33, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.

1 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.

OpenAI logo
Model A
GPT-6 Luna

OpenAI

62.33/100

Estimated · Public rank #56

90% interval 45.7–73.8

xAI logo
Model B
Grok 4.20

xAI

67.97/100

Supported · Public rank #32

90% interval 59.4–76.5

Updated September 22, 2026. Rank says Grok 4.20 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

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

    Grok 4.20

    Grok 4.20 has the larger documented context window.

    Confidence: documented

  • Chat turn cost

    1K fresh input + 500 output tokens

    GPT-6 Luna

    GPT-6 Luna 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-6 Luna

    GPT-6 Luna has the lower estimated token cost for this stated workload. Grok 4.20 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

    GPT-6 Luna

    GPT-6 Luna 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

    GPT-6 Luna 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

    GPT-6 Luna 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.

66.6GPT-6 Luna29.8Grok 4.20

Directional only · BenchAlign

GPT-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 leaderboardShortlist models for this job

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
1
GPT-6 Luna only
6
Grok 4.20 only
22
Like-for-like categories
0 / 8

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

Category results, on a stated basis

Each row shows the public-lane category score for both models: the BenchAlign 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-6 Luna
56.7
Estimated · #35/157
Grok 4.20
29.5
Supported · #147/157
Basis
BenchAlign lane · 1 vs 4 public rows
Reading
Directional only

Coding

Directional only
GPT-6 Luna
66.6
Estimated · #11/159
Grok 4.20
29.8
Supported · #149/159
Basis
BenchAlign lane · 1 vs 6 public rows
Reading
Directional only

Knowledge

Directional only
GPT-6 Luna
63.6
Estimated · #29/189
Grok 4.20
50.5
Supported · #87/189
Basis
BenchAlign lane · 5 vs 6 public rows
Reading
Directional only

Reasoning

Not comparable
GPT-6 Luna
78.3
Unranked · 2 rankable rows
Grok 4.20
34.3
Unranked · 2 rankable rows
Basis
Provisional lane · 0 vs 2 weighted rows
Reading
Not comparable

Multimodal

Not comparable
GPT-6 Luna
71.3
Unranked · 1 rankable row
Grok 4.20
34.6
#45/50
Basis
Provisional lane · 0 vs 2 weighted rows
Reading
Not comparable

Multilingual

Not comparable
GPT-6 Luna
Not ranked
Grok 4.20
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Instruction following

Not comparable
GPT-6 Luna
Not ranked
Grok 4.20
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Math

Not comparable
GPT-6 Luna
Not ranked
Grok 4.20
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

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

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.

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-6 Luna
$0.00035
Fits in one request
Grok 4.20
$0.005
Fits in one request

GPT-6 Luna has the lower modeled cost

Costs use the listed standard API rates.

Repository review

50K fresh input + 3K output tokens

GPT-6 Luna
$0.0065
Fits in one request
Grok 4.20
$0.118
Fits in one request

GPT-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-6 Luna
$0.009
Fits in one request
Grok 4.20
$0.5
Fits in one request
Cached input priced at the published list-input rate

GPT-6 Luna has the lower modeled cost

Grok 4.20 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-6 Luna

$0.01 per 1M cached input tokens

OpenAI GPT-6 Luna model documentation

Grok 4.20

Not published

Provider availability

GPT-6 Luna

Generally Available · OpenAI Responses API, OpenAI Chat Completions API, ChatGPT Work, Codex

OpenAI GPT-6 Sol and Luna launch

Grok 4.20

Not sourced

Reasoning profile

GPT-6 Luna

Reasoning

Grok 4.20

Reasoning

Weight access

GPT-6 Luna

Proprietary

Grok 4.20

Proprietary

License

GPT-6 Luna

Proprietary

Grok 4.20

Proprietary

Release date

GPT-6 Luna

2026-09-16

Grok 4.20

2026-03-10

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
Grok 4.20 has the higher public score estimate, 67.97 versus 62.33, but the 90% score intervals overlap.
Workload cost
Repository review: $0.0065 vs $0.118. Cache-heavy agent loop: $0.009 vs $0.5.
Context tradeoff
Grok 4.20 has the larger documented window (2M).

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 evidence29 rows

Agentic

  • ExploitGym

    GPT-6 Luna11.6%
    Source
    Grok 4.20—

    Not directly comparable

  • Terminal-Bench 2.0

    GPT-6 Luna—
    Grok 4.2047.1%
    Source

    Not directly comparable

  • DeepSearchQA

    GPT-6 Luna—
    Grok 4.2062.8%
    Source

    Not directly comparable

  • Gert Labs

    GPT-6 Luna—
    Grok 4.2038.36%
    Source

    Not directly comparable

  • Terminal-Bench 2.1 (Vals)

    GPT-6 Luna—
    Grok 4.2044.2%
    Source

    Not directly comparable

Coding

  • DeepSWE

    GPT-6 Luna66.6%
    Source
    Grok 4.20—

    Not directly comparable

  • LiveCodeBench Pro

    GPT-6 Luna—
    Grok 4.2074.2%
    Source

    Not directly comparable

  • SWE-bench Verified

    GPT-6 Luna—
    Grok 4.2076.7%
    Source

    Not directly comparable

  • SWE-bench Pro

    GPT-6 Luna—
    Grok 4.2051.8%
    Source

    Not directly comparable

  • Vibe Code Bench

    GPT-6 Luna—
    Grok 4.204.06%
    Source

    Not directly comparable

  • LiveCodeBench (Vals)

    GPT-6 Luna—
    Grok 4.2084.3%
    Source

    Not directly comparable

  • SWE-bench (Vals)

    GPT-6 Luna—
    Grok 4.2072.2%
    Source

    Not directly comparable

Reasoning

  • ARC-AGI-2

    GPT-6 Luna—
    Grok 4.2053.3%
    Source

    Not directly comparable

  • ARC-AGI-3

    GPT-6 Luna—
    Grok 4.200.1%
    Source

    Not directly comparable

Multimodal

  • MMMU-Pro

    GPT-6 Luna—
    Grok 4.2075.2%
    Source

    Not directly comparable

  • CharXiv

    GPT-6 Luna—
    Grok 4.2060.9%
    Source

    Not directly comparable

  • ERQA

    GPT-6 Luna—
    Grok 4.2054.1%
    Source

    Not directly comparable

  • SimpleVQA

    GPT-6 Luna—
    Grok 4.2057.4%
    Source

    Not directly comparable

  • MedXpertQA (MM)

    GPT-6 Luna—
    Grok 4.2065.8%
    Source

    Not directly comparable

Knowledge

  • HealthBench (raw)

    GPT-6 Luna50.0%
    Source
    Grok 4.20—

    Not directly comparable

  • HealthBench (length-adjusted)

    GPT-6 Luna54.5%
    Source
    Grok 4.20—

    Not directly comparable

  • HealthBench Professional

    GPT-6 Luna60.8%
    Source
    Grok 4.20—

    Not directly comparable

  • HealthBench Professional (raw)

    GPT-6 Luna61.2%
    Source
    Grok 4.20—

    Not directly comparable

  • HealthBench Hard

    GPT-6 Luna31.4%
    Source
    Grok 4.2020.3%
    Source

    GPT-6 Luna leads this result

  • GPQA-D

    GPT-6 Luna—
    Grok 4.2088.5%
    Source

    Not directly comparable

  • HLE w/o tools

    GPT-6 Luna—
    Grok 4.2031.6%
    Source

    Not directly comparable

  • MedXpertQA (Text)

    GPT-6 Luna—
    Grok 4.2050.2%
    Source

    Not directly comparable

  • GPQA Diamond (Vals)

    GPT-6 Luna—
    Grok 4.2088.6%
    Source

    Not directly comparable

  • MMLU-Pro (Vals)

    GPT-6 Luna—
    Grok 4.2086.3%
    Source

    Not directly comparable

Questions

Which is better, GPT-6 Luna or Grok 4.20?

Grok 4.20 has the higher public score estimate, 67.97 versus 62.33, 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-6 Luna or Grok 4.20?

GPT-6 Luna scores higher for coding on the public lane, 66.6 to 29.8. GPT-6 Luna 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-6 Luna or Grok 4.20?

GPT-6 Luna scores higher for agentic tasks on the public lane, 56.7 to 29.5. GPT-6 Luna 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-6 Luna or Grok 4.20?

For the stated presets, chat costs $0.00035 on GPT-6 Luna and $0.005 on Grok 4.20; repository review costs $0.0065 and $0.118; the cache-heavy agent loop costs $0.009 and $0.5. Grok 4.20 has no published cached-input rate, so cached tokens use its listed input rate.

Which has the larger context window, GPT-6 Luna or Grok 4.20?

Grok 4.20 has the larger documented context window: 2M, compared with 1.05M.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated September 22, 2026

Watch GPT-6 Luna vs Grok 4.20

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

Read a sample issue

Join 2,000+ readers.