Gemma 3 4B vs GPT-5.2-Codex
GPT-5.2-Codex leads Gemma 3 4B on the Artificial Analysis Intelligence Index, 28.5 to 4.8, and Gemma 3 4B costs 186.7x less per output token ($0.075 vs $14.00 per million). Updated October 21, 2026; 7 benchmarks have a result for both.
Gemma 3 4B or GPT-5.2-Codex: which to pick
Only GPT-5.2-Codex has a coding score, 45.4 on the Vibe Coding Index; Gemma 3 4B has none, so the two cannot be ranked on coding.
Pick Gemma 3 4B for output tokens 186.7x cheaper ($0.075 vs $14.00 per million).
- Shared results
- 7
- Gemma 3 4B only
- 32
- GPT-5.2-Codex only
- 15
- Like-for-like categories
- 5 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Only GPT-5.2-Codex has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
GPT-5.2-Codex
GPT-5.2-Codex leads, 28.5 against 4.8.
Same board, both measured
Long documents
Listed context window
GPT-5.2-Codex
GPT-5.2-Codex takes 400K tokens against 131K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemma 3 4B
Gemma 3 4B $0.0001, GPT-5.2-Codex $0.0088 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5.2-Codex has the number, $0.43.
Directional only
At a glance
- Coding index
- Gemma 3 4B not benchmarked, GPT-5.2-Codex 45.4 from 3 benchmarks.
- Widest category
- Agentic work, 87.1 points apart on average over 1 shared benchmark.
- Widest single gap
- τ²-Bench Telecom: Gemma 3 4B 5.0% against GPT-5.2-Codex 92.1%, 87.1 points apart on a 0 to 100 scale.
- Output price
- Gemma 3 4B $0.075 against GPT-5.2-Codex $14.00 per million output tokens, 186.7x apart.
- Context window
- Gemma 3 4B takes 131K tokens, GPT-5.2-Codex 400K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Gemma 3 4B | Bars | GPT-5.2-Codex | Ties |
|---|---|---|---|---|
| Coding | 0 | 1 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 0 | 2 | ||
| General | 0 | 2 | ||
| All shared boards | 0 | 7 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index01
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| SWE-bench Verified | Not measured | 72.8% | Only GPT-5.2-Codex measured |
| SWE-bench Multilingual | Not measured | 66.3% | Only GPT-5.2-Codex measured |
| LMArena WebDev | Not measured | 1339 | Only GPT-5.2-Codex measured |
| AA Coding Index | 2.7 | Not measured | Only Gemma 3 4B measured |
| LiveCodeBench (Vals AI) | Not measured | 88.0% | Only GPT-5.2-Codex measured |
| SWE-bench Verified (Vals AI) | Not measured | 72.4% | Only GPT-5.2-Codex measured |
| Vibe Code Bench (Vals AI) | Not measured | 37.9% | Only GPT-5.2-Codex measured |
| LiveCodeBench (Artificial Analysis run) | 11.2% | Not measured | Only Gemma 3 4B measured |
| Terminal-Bench Hard (Artificial Analysis run) | 0.8% | 37.1% | GPT-5.2-Codex |
| Terminal-Bench | Not measured | 66.5% | Only GPT-5.2-Codex measured |
| ALE-Bench | Not measured | 1299.9 | Only GPT-5.2-Codex measured |
| LMArena Text: coding | 1230 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: software and IT | 1269 | Not measured | Only Gemma 3 4B measured |
| LMArena WebDev: React | Not measured | 1329 | Only GPT-5.2-Codex measured |
| LMArena WebDev: HTML | Not measured | 1348 | Only GPT-5.2-Codex measured |
| Neither model has a result on the other 23 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| AA Intelligence Index | 4.8 | 28.5 | GPT-5.2-Codex |
| Epoch Capabilities Index | 115.9 | Not measured | Only Gemma 3 4B measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | Not measured | 1051 | Only GPT-5.2-Codex measured |
| Design Arena: agents building full-stack apps | Not measured | 993 | Only GPT-5.2-Codex measured |
| Design Arena: agents building mobile apps | Not measured | 1090 | Only GPT-5.2-Codex measured |
| Design Arena: agents building Android apps | Not measured | 1176 | Only GPT-5.2-Codex measured |
| Design Arena: agents building Godot games | Not measured | 1142 | Only GPT-5.2-Codex measured |
| τ²-Bench Telecom | 5.0% | 92.1% | GPT-5.2-Codex |
| Neither model has a result on the other 33 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 29.1% | 89.9% | GPT-5.2-Codex |
| Humanity's Last Exam (Artificial Analysis run) | 5.3% | 35.7% | GPT-5.2-Codex |
| GPQA Diamond (Epoch AI run) | 23.2% | Not measured | Only Gemma 3 4B measured |
| LMArena Text: hard prompts | 1253 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: expert | 1223 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: hard prompts in English | 1253 | Not measured | Only Gemma 3 4B measured |
| Neither model has a result on the other 15 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| AA Math Index | 12.7 | Not measured | Only Gemma 3 4B measured |
| AIME 2025 (Artificial Analysis run) | 12.7% | Not measured | Only Gemma 3 4B measured |
| OTIS Mock AIME | 7.5% | Not measured | Only Gemma 3 4B measured |
| LMArena Text: math | 1239 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: mathematical work | 1256 | Not measured | Only Gemma 3 4B measured |
| Neither model has a result on the other 8 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | 41.7% | Not measured | Only Gemma 3 4B measured |
| LMArena Text: science | 1306 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: medicine and healthcare | 1342 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: legal and government | 1327 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: business and finance | 1311 | Not measured | Only Gemma 3 4B measured |
| Neither model has a result on the other 12 boards here | |||
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| Neither model has a result on 9 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| IFBench | 28.3% | 77.6% | GPT-5.2-Codex |
| Long context reasoning (AA-LCR) | 6.7% | 82.3% | GPT-5.2-Codex |
| DTBench | 50.9% | Not measured | Only Gemma 3 4B measured |
| LMCA | 2.8% | Not measured | Only Gemma 3 4B measured |
| LMArena Text | 1291 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: instruction following | 1239 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: creative writing | 1270 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: English | 1300 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: non-English | 1273 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: German | 1281 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: Russian | 1294 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: multi-turn | 1255 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: longer queries | 1273 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: writing and language | 1288 | Not measured | Only Gemma 3 4B measured |
| LMArena Text: entertainment and media | 1233 | Not measured | Only Gemma 3 4B measured |
| Neither model has a result on the other 31 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Gemma 3 4B | GPT-5.2-Codex | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Gemma 3 4B costs less
Repository review
50K input + 3K output tokens
Gemma 3 4B costs less
Agent loop
200K input + 10K output tokens
Gemma 3 4B costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Google Gemma 3 4B | OpenAI GPT-5.2-Codex |
|---|---|---|
| Maker | OpenAI | |
| Weights | Open weights | No public weights listed |
| Released | March 13, 2025 | January 14, 2026 |
| Context window | 131K | 400K |
| Input $/Mtok | $0.045 | $1.75 |
| Output $/Mtok | $0.075 | $14.00 |
| Coding index | Not benchmarked | 45.4 |
| Independent measurements behind it | 0 | 3 |
| Cost per solved task | Not computable | $0.43 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Gemma 3 4B better than GPT-5.2-Codex for coding?
Only GPT-5.2-Codex has a coding score, 45.4 on the Vibe Coding Index; Gemma 3 4B has none, so the two cannot be ranked on coding. Pick Gemma 3 4B for output tokens 186.7x cheaper ($0.075 vs $14.00 per million).
Which is cheaper, Gemma 3 4B or GPT-5.2-Codex?
Gemma 3 4B costs $0.045 per million input tokens and $0.075 per million output tokens; GPT-5.2-Codex costs $1.75 and $14.00. Cost per solved task is known for GPT-5.2-Codex only.
Which has the bigger context window, Gemma 3 4B or GPT-5.2-Codex?
GPT-5.2-Codex, with 400K tokens against 131K.
How many benchmarks is the Gemma 3 4B vs GPT-5.2-Codex comparison based on?
7 benchmarks have a result for both models as of October 21, 2026. Gemma 3 4B's coding index rests on 0 independent benchmarks and GPT-5.2-Codex's on 3. Where only one model has a result, the page says so instead of naming a winner.