Models / Gemma 4 12B

Google

Gemma 4 12B

Gemma 4 12B trails most of the field on what it has been measured on; it is not on sale through OpenRouter, so it is here for reference.

Results 16 Nov 2026 · 26 results · 2 sources

Price per million tokens
Not available through OpenRouter
Speed
148 tokens a second; on its test prompt the answer starts after 14.7 s, thinking time included
Artificial Analysis, on its usual API; 145–148 across 2 settings
Intelligence Index
14.2
Artificial Analysis; 9.4–14.2 across 2 settings
Spring Prompt overall
Not ranked yet
needs our benchmarks and two groups of results
Developer
Google, based in the US
Weights not published
Released
3 Jun 2026 (Artificial Analysis)

Where it stands among the models you could choose

Available models grouped by price, intelligence, speed and developer location. Retired models are excluded.

AmongIntelligence IndexSpring Prompt overallPrice, blendedSpeed
Models available today 115th of 186Claude Opus 5.5 – – 22nd of 66Nemotron 3.5 Super VL
Models with similar intelligence sets this group – – 5th of 11gpt-oss-120b
Models writing 100–200 tokens a second 17th of 22Claude Sonnet 5.5 – – sets this group
Models from developers based in the US 63rd of 93Claude Opus 5.5 – – 14th of 34Nemotron 3.5 Super VL

The name under each rank is the leader of that group; green is the top quarter of the group and red the bottom quarter, by rank. Price is shown as a percentile: the share of the group that costs less, so lower is cheaper. "Sets this group" marks the measure the group is defined by. Spring Prompt overall is our score out of 100 across every source (how it works). Price is the developer's list price (or the typical OpenRouter host where we haven't read one) for three input tokens to one output; intelligence and speed are from Artificial Analysis (data sourced from Artificial Analysis); "similar intelligence" means within 4 points on its Intelligence Index. Developer locations are where each company is based, not where a model is served. Groups under 3 models are not ranked.

How good is it, and for what?

Each use case ranks the available models its benchmarks measured. The bar shows its percentile in that field, best to the right. Open a row for the results behind it.

Product listingsTurning a sparse product feed and photos into listings that can go live Not measured
  • CatalogBench: has not measured this model.
Decks from an analysisTurning a finished analysis into a deck you could present as it is Not measured
  • DeckBench: has not measured this model.
User surveysPlanning a user survey and reading its results without being misled Not measured
  • SurveyBench: has not measured this model.
Marketing planningPlanning a year of ad spend without overspending Not measured
  • ROASBench: has not measured this model.
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 115th of 160 available
BenchmarkGemma 4 12BRank among availableBest available
Artificial AnalysisArtificial Analysis: Terminal-Bench 2.1 · reported 27.3% 79th of 108 Claude Fable 5.1 91.4%
  • tau2-bench: pending. Its latest results are dated 26 May 2026, before this model was released.
  • Berkeley Function Calling Leaderboard (BFCL) V4: pending. Its latest results are dated 16 Dec 2025, before this model was released.
  • OpenHands Index: has not measured this model.
  • Microsoft STATE-Bench: pending. Its latest results are dated 29 May 2026, before this model was released.
  • Vending-Bench 2: has not measured this model.
Professional workReal tasks from banking, consulting and law, business documents and freelance projects Not measured
  • APEX-Agents: has not measured this model.
  • GDP.pdf: has not measured this model.
  • Remote Labor Index: has not measured this model.
Reasoning and knowledgeHard questions across science, maths and general knowledge 121st of 188 available
BenchmarkGemma 4 12BRank among availableBest available
Artificial AnalysisArtificial Analysis: Artificial Analysis Intelligence Index · reported 14.2 115th of 186 Claude Opus 5.5 57.6
Artificial AnalysisArtificial Analysis: Humanity's Last Exam · reported 15.7% 114th of 187 Claude Opus 5.5 61.4%
Artificial AnalysisArtificial Analysis: GPQA Diamond · reported 75.3% 108th of 169 GPT-6 Astra 96.3%
WritingWhat people prefer in blind comparisons, and judged writing quality 118th of 170 available
BenchmarkGemma 4 12BRank among availableBest available
UGI LeaderboardUGI: writing score · reported 36.3 88th of 125 Gemini 3.8 Flash 78.6
  • Arena (formerly LMArena): has not measured this model.
Sticking to the factsSummarising without inventing things, and factual answers Not measured
  • Vectara Hallucination Leaderboard: has not measured this model.
  • Arena (formerly LMArena): has not measured this model.
  • SimpleQA Verified (Epoch AI): has not measured this model.
Speed under pressureGood decisions against a real clock (fast chess) Not measured
  • BulletBench: has not measured this model.

Nearest alternatives

  • Google's newer, at least as capable modelGemini 4 Argon$4.00 blended · Intelligence Index 52.6
  • Open weights at similar intelligenceGranite 4.2 8B$0.08 blended · Intelligence Index 11.1

Against Gemma 4 12B, Intelligence Index 14.2. "Similar intelligence" means within 4 points or better.

Against the alternatives

The models you would most likely weigh it against: the leaders of the groups above.

Scroll sideways to see every alternative.

Gemma 4 12Bgpt-oss-120b
fastest with similar intelligence
Claude Opus 5.5
most intelligent available today
Claude Fable 5.2
leads overall
Gemini 4 Argon
Google's best other model
Price per million tokens – $0.20$8.00$16.00$4.00
Tokens a second 148 25996––
Intelligence Index 14.2 11.657.6–52.6
Spring Prompt overall – –829072
Rank among available models, by use case
Product listings – –13th5th11th
Decks from an analysis – –7th4th23rd
User surveys – –4th1st15th
Marketing planning – –9th1st13th
Agents and tool use 115th of 160 126th3rd–5th
Professional work – 49th3rd––
Reasoning and knowledge 121st of 188 118th1st–4th
Writing 118th of 170 123rd5th–1st
Sticking to the facts – 129th3rd–1st
Speed under pressure – ––8th26th

Shaded figures are better than Gemma 4 12B's. A dash means no result.

Where it has been measured

Compared with the models available today (it is not on sale itself), mid-field on every use case measured so far.

Every result

26 published results from 2 sources, each in the source's own units, with the configuration that produced it.

Artificial Analysis · reported by Artificial Analysis · 20 results
MeasureValueRankConfigurationDated
Artificial Analysis: Artificial Analysis Coding Indexindex score, higher is better 31.0 135 of 187 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 14.2 262 of 413 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 9.4 330 of 413 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 75.3% 214 of 360 Gemma 4 12B 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 66.1% 274 of 360 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 15.7% 230 of 419 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 6.3% 326 of 419 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 73.5% 35 of 288 Gemma 4 12B 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 45.2% 167 of 288 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 63.7% 240 of 407 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 35.0% 326 of 407 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: Output speedtokens per second, higher is better 148 38 of 137 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Output speedtokens per second, higher is better 145 40 of 137 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench 2.1% of tasks, higher is better 27.3% 138 of 188 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 18.2% 155 of 282 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 11.4% 194 of 282 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: Time to first answer tokenseconds, lower is better 14.7 75 of 137 Gemma 4 12B 20 Oct 2026
Artificial Analysis: Time to first answer tokenseconds, lower is better 1.15 30 of 137 Gemma 4 12B (no reasoning) 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 36.3% 189 of 286 Gemma 4 12B 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 31.9% 204 of 286 Gemma 4 12B (no reasoning) 20 Oct 2026

Not shown: Not your codebase or tools.

UGI Leaderboard · reported by UGI Leaderboard · 6 results
MeasureValueRankConfigurationDated
UGI: requested-length error% off the requested word count, lower is better 8.0% 63 of 379 Gemma 4 12B 9 Jun 2026
UGI: requested-length error% off the requested word count, lower is better 8.0% 63 of 379 Gemma 4 12B (channel-thought-prefill reasoning) 9 Jun 2026
UGI: style adherencescore from 0 to 1, higher is better 0.35 210 of 379 Gemma 4 12B 9 Jun 2026
UGI: style adherencescore from 0 to 1, higher is better 0.33 250 of 379 Gemma 4 12B (channel-thought-prefill reasoning) 9 Jun 2026
UGI: writing scorescore out of 100, higher is better 31.6 271 of 379 Gemma 4 12B 9 Jun 2026
UGI: writing scorescore out of 100, higher is better 36.3 241 of 379 Gemma 4 12B (channel-thought-prefill reasoning) 9 Jun 2026

Not shown: Not other format limits such as character counts or bullet counts.

Test your own prompt.

Benchmarks aren't your task. Run your prompt on the models you could use, with every answer checked and costed.