GPT-5.6 Luna vs Grok4.5
Wins 2 of 8 areas
Agents · Math
Wins 5 of 8 areas
Coding · Reasoning · Facts · Images and charts · Following instructions
Grok4.5 is the stronger all-rounder.GPT-5.6 Luna is cheaper and better at agents.
Scores updated · 41 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software14Grok4.54 of 6 tests · 1 tie
- FactsGetting facts right instead of making them up03Grok4.53 of 3 tests
- ReasoningHard problems that need careful thinking24Grok4.54 of 7 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video12Grok4.52 of 3 tests
- Following instructionsDoing exactly what it is asked01Grok4.51 of 1 test
- AgentsCarrying out multi-step tasks on its own21GPT-5.6 Luna2 of 3 tests
- MathCompetition and research-level math21GPT-5.6 Luna2 of 3 tests
- Long documentsFinding answers in very long texts11Even1 each
Following instructions rests on a single test.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
GPT-5.6 Luna costs 83% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Grok4.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+38.5points ahead
- Common-sense trick questionsSimpleBench+23.2points ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+11.4points ahead
Where GPT-5.6 Luna pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+34.1points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+24.9points ahead
- Programming problems, refreshed regularlyLiveBench · Coding+14.3points ahead
Every test, side by side
All 41 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGrok4.5
- LiveBench · CodingGPT-5.6 Luna by 14.382.968.6+14.3
- LiveBench · Agentic CodingGrok4.5 by 8.148.456.5+8.1
- LMArena · WebDevGrok4.5 by 34 rating points15181552+34 rating
- SWE-bench ProGrok4.5 by 262.764.7+2
- SciCodeGrok4.5 by 1.453.655+1.4
- Terminal-Bench 2.1tie80.981.6tie
AgentsGPT-5.6 Luna
- τ-Bench V3 · BankingGrok4.5 by 1131.142.1+11
- GDPValGPT-5.6 Luna by 3.748.244.5+3.7
- Terminal-Bench 4.0GPT-5.6 Luna by 111.610.6+1
ReasoningGrok4.5
- SimpleBenchGrok4.5 by 23.246.870+23.2
- ARC-AGI-2GPT-5.6 Luna by 759.652.6+7
- CritPtGPT-5.6 Luna by 5.220.615.4+5.2
- Humanity's Last ExamGrok4.5 by 3.239.542.7+3.2
- GPQA DiamondGrok4.5 by 291.193.1+2
- LiveBench · ReasoningGrok4.5 by 1.685.687.2+1.6
- ARC-AGI-3tie0.20.3tie
FactsGrok4.5
- AA-Omniscience · Non-hallucinationGrok4.5 by 38.57.445.9+38.5
- AA-Omniscience · AccuracyGrok4.5 by 8.942.751.5+8.9
- SimpleQA VerifiedGrok4.5 by 7.34148.3+7.3
Images and chartsGrok4.5
- LMArena · VisionGrok4.5 by 28 rating points12601288+28 rating
- MMMU-ProGrok4.5 by 1.878.680.4+1.8
- CharXiv (RQ)GPT-5.6 Luna by 1.182.781.6+1.1
MathGPT-5.6 Luna
- FrontierMath Tier 4GPT-5.6 Luna by 34.158.524.4+34.1
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Luna by 24.982.157.2+24.9
- LiveBench · MathematicsGrok4.5 by 3.687.290.8+3.6
Long documentsEven
- MRCR v2 8 needle 128k (average)Grok4.5 by 6.674.881.4+6.6
- AA-LCRGPT-5.6 Luna by 4.483.779.3+4.4
Following instructionsGrok4.5
- LiveBench · Instruction FollowingGrok4.5 by 11.460.171.5+11.4
Other results13 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- AA-OmniscienceGrok4.5 by 35.6-10.325.3+35.6
- DeepSWEGPT-5.6 Luna by 14.267.253+14.2
- DeepSWE 1.1GPT-5.6 Luna by 136754+13
- livebench_languageGrok4.5 by 10.272.682.8+10.2
- τ³-Bench BankingGrok4.5 by 8.724.333+8.7
- ARC-AGI-1GPT-5.6 Luna by 590.785.7+5
- livebench_data_analysisGPT-5.6 Luna by 57873+5
- GDPval-AA v2 EloGPT-5.6 Luna by 49 rating points15841535+49 rating
- Terminal-Bench 4.0GPT-5.6 Luna by 4.917.312.4+4.9
- MLE-BenchGPT-5.6 Luna by 4.447.643.2+4.4
- AA IntelligenceGrok4.5 by 1.537.338.8+1.5
- Artificial Analysis Coding IndexGrok4.5 by 171.472.4+1
- AA Agentic Indextie42.742.1tie
Questions people ask
Which is better, GPT-5.6 Luna or Grok4.5?
Grok4.5 wins five of the eight areas we test: coding, reasoning, facts, images and charts and following instructions. GPT-5.6 Luna wins agents and math, and costs 83% less. They are level on long documents.
Which is better for coding?
Grok4.5. It wins 4 of the 6 coding tests both models report; GPT-5.6 Luna wins 1, and 1 is a tie.
Which is cheaper?
GPT-5.6 Luna costs $0.20 per million input tokens and $1.20 per million output tokens; Grok4.5 costs $2.00 and $6.00. That makes GPT-5.6 Luna about 83% cheaper for the same work.
How do you compare the two?
We use the 41 benchmark tests both models have published scores on. The verdict counts the 28 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 13 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.