Models / Gemma 4 31B

Google

Gemma 4 31B

Gemma 4 31B is strongest at writing; worth a look for simple, high-volume work where cost matters most. The cheapest model of similar intelligence is Granite 4.2 8B, at $0.08 per million tokens against $0.21.

Results 16 Nov 2026 · 43 results · 5 sources

Test Gemma 4 31B on your prompt → It runs beside the models you could use instead, with every answer checked and costed.

Price per million tokens
$0.15 in · $0.40 out
Typical host on OpenRouter, 10 Nov 2026 · compare prices
Speed
48 tokens a second; on its test prompt the answer starts after 0.8 s, thinking time included
Artificial Analysis, on its usual API, at no reasoning; 35–48 across 2 settings
Intelligence Index
14.7
Artificial Analysis; 13.9–14.7 across 2 settings
Spring Prompt overall
Not ranked yet
needs our benchmarks and two groups of results
Developer
Google, based in the US
Weights published
Released
2 Apr 2026
262,144 tokens of context

Where it stands among the models you could choose

Available models grouped by price, intelligence, speed and developer location, plus models with published weights. Retired models are excluded.

AmongIntelligence IndexSpring Prompt overallPrice, blendedSpeed
Models available today 114th of 185Claude Opus 5.5 – $0.21 blended, 21st percentilecheapest: Mistral Nemo 57th of 65Nemotron 3.5 Super VL
Models priced under $0.25 per million tokens 11th of 41Claude Haiku 5.5 – sets this group 17th of 19Nemotron 3.5 Super VL
Models with similar intelligence sets this group – $0.21 blended, 26th percentilecheapest: Granite 4.2 8B 10th of 10gpt-oss-120b
Models writing under 50 tokens a second 7th of 9MiMo-V2.6-Pro – $0.21 blended, 25th percentilecheapest: Phi-4 sets this group
Models from developers based in the US 62nd of 92Claude Opus 5.5 – $0.21 blended, 22nd percentilecheapest: Gemma 3 4B 31st of 33Nemotron 3.5 Super VL
Open-weights models 51st of 104MiMo-V2.6-Pro – $0.21 blended, 29th percentilecheapest: Mistral Nemo 36th of 43Nemotron 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, then those priced under $0.25 per million tokens. 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 106th of 159 available · 12th of 37 at its price
BenchmarkGemma 4 31BRank among availableBest available
Artificial AnalysisArtificial Analysis: Terminal-Bench 4.0 · reported 0.0% 67th of 107 Kimi K3.1 Pro 67.2%
Artificial AnalysisArtificial Analysis: Terminal-Bench 2.1 · reported 43.4% 68th of 107 Claude Fable 5.1 91.4%
Artificial AnalysisArtificial Analysis: τ-bench banking · reported 14.8% 57th of 104 Grok 4.6 50.7%
  • tau2-bench: has not measured this model.
  • 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: has not measured this model.
  • 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 99th of 187 available · 8th of 41 at its price
BenchmarkGemma 4 31BRank among availableBest available
Artificial AnalysisArtificial Analysis: Artificial Analysis Intelligence Index · reported 14.7 114th of 185 Claude Opus 5.5 57.6
Artificial AnalysisArtificial Analysis: Humanity's Last Exam · reported 23.6% 94th of 186 Claude Opus 5.5 61.4%
Artificial AnalysisArtificial Analysis: GPQA Diamond · reported 85.7% 67th of 168 GPT-6 Astra 96.3%
WritingWhat people prefer in blind comparisons, and judged writing quality 72nd of 169 available · 5th of 39 at its price
BenchmarkGemma 4 31BRank among availableBest available
Arena (formerly LMArena)Arena Text: overall · reported 1,453 51st of 153 Gemini 4 Argon 1,525
UGI LeaderboardUGI: writing score · reported 43.1 69th of 124 Gemini 3.8 Flash 78.6
Sticking to the factsSummarising without inventing things, and factual answers 85th of 139 available · 15th of 30 at its price
BenchmarkGemma 4 31BRank among availableBest available
Vectara Hallucination LeaderboardVectara: hallucination rate · reported 7.4% 24th of 71 GPT-5.4 nano 3.1%
Arena (formerly LMArena)Arena Text factuality: overall · reported 1,450 49th of 109 Gemini 4 Argon 1,511
SimpleQA Verified (Epoch AI)SimpleQA Verified: correct answers · reported 10.4% 66th of 68 GPT-6 Astra 75.6%
Speed under pressureGood decisions against a real clock (fast chess) Not measured
  • BulletBench: has not measured this model.

Nearest alternatives

Against Gemma 4 31B at $0.21 blended, Intelligence Index 14.7. "Similar intelligence" means within 4 points or better.

Where to run it

17 hosts on OpenRouter at the standard service tier, with a median of $0.21 blended per million tokens; quantisation where the host states it: FP4, BF16, FP8. Cheapest first.

HostQuantisationInputOutputBlended
CoreWeaveFP4$0.10$0.34$0.16
VeniceFP4$0.12$0.36$0.18
ChutesFP4$0.12$0.37$0.18
CrusoeBF16$0.14$0.40$0.21
DekaLLMnot stated$0.14$0.40$0.21
Friendlinot stated$0.14$0.40$0.21
NovitaBF16$0.14$0.40$0.21
NextBitFP4$0.15$0.40$0.21

Other service tiers, not counted above: DeepInfra flex $0.16 in, $0.32 out; DeepInfra priority $0.30 in, $0.60 out. Flex and batch are cheaper but slower or queued; priority and fast cost more.

Prices per million tokens from OpenRouter's public endpoint list, checked 10 Nov 2026; the 9 dearer hosts are left out.

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 31BClaude Haiku 5.5
most intelligent at under $0.25 per million tokens
Granite 4.2 8B
cheapest with similar intelligence
gpt-oss-120b
fastest with similar intelligence
Claude Opus 5.5
most intelligent available today
Price per million tokens $0.21 $0.20$0.08$0.20$8.00
Tokens a second 48 2406525996
Intelligence Index 14.7 43.411.111.657.6
Spring Prompt overall – 40––82
Rank among available models, by use case
Product listings – 20th––13th
Decks from an analysis – 18th––7th
User surveys – 7th––4th
Marketing planning – 30th––9th
Agents and tool use 106th of 159 28th126th125th3rd
Professional work – ––49th3rd
Reasoning and knowledge 99th of 187 25th138th118th1st
Writing 72nd of 169 –158th124th5th
Sticking to the facts 85th of 139 –137th129th3rd
Speed under pressure – 21st–––

Shaded figures are better than Gemma 4 31B's. A dash means no result.

Where it has been measured

3rd cheapest of the 9 models writing under 50 tokens a second. 11th most intelligent of the 41 models priced under $0.25 per million tokens. Among models available today, mid-field on every use case measured so far.

Every result

43 published results from 5 sources, each in the source's own units, with the configuration that produced it.

Arena (formerly LMArena) · reported by Arena (formerly LMArena) · 8 results
MeasureValueRankConfigurationDated
Arena Document: overallArena rating, higher is better 1,445
1,437–1,452
34 of 44 Gemma 4 31B 13 Sep 2026
Arena Text factuality: overallArena rating, higher is better 1,450
1,443–1,457
69 of 182 Gemma 4 31B 2 Oct 2026
Arena Text: business, management and financeArena rating, higher is better 1,449
1,432–1,466
76 of 406 Gemma 4 31B 2 Oct 2026
Arena Text: creative writingArena rating, higher is better 1,420
1,401–1,439
86 of 411 Gemma 4 31B 2 Oct 2026
Arena Text: expert promptsArena rating, higher is better 1,498
1,473–1,523
48 of 363 Gemma 4 31B 2 Oct 2026
Arena Text: instruction followingArena rating, higher is better 1,453
1,439–1,467
59 of 413 Gemma 4 31B 2 Oct 2026
Arena Text: overallArena rating, higher is better 1,453
1,445–1,460
76 of 413 Gemma 4 31B 2 Oct 2026
Arena Text: writing, literature and languageArena rating, higher is better 1,435
1,420–1,451
75 of 412 Gemma 4 31B 2 Oct 2026

Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Artificial Analysis · reported by Artificial Analysis · 26 results
MeasureValueRankConfigurationDated
Artificial Analysis: Artificial Analysis Coding Indexindex score, higher is better 43.4 116 of 187 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Artificial Analysis Coding Indexindex score, higher is better 33.2 132 of 187 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 14.7 259 of 413 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 13.9 267 of 413 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 85.7% 118 of 360 Gemma 4 31B 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 76.3% 209 of 360 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 23.6% 185 of 419 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 11.8% 260 of 419 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 75.6% 26 of 288 Gemma 4 31B 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 53.5% 127 of 288 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 69.7% 204 of 407 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 46.7% 289 of 407 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: Output speedtokens per second, higher is better 35.4 133 of 137 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Output speedtokens per second, higher is better 47.6 118 of 137 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: SciCode% of problems, higher is better 45.5% 144 of 205 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Terminal-Bench 2.1% of tasks, higher is better 43.4% 119 of 188 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Terminal-Bench 2.1% of tasks, higher is better 29.2% 135 of 188 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench 4.0% of tasks, higher is better 0.0% 152 of 204 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 36.4% 69 of 282 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 30.3% 107 of 282 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: Time to first answer tokenseconds, lower is better 50.0 116 of 137 Gemma 4 31B 20 Oct 2026
Artificial Analysis: Time to first answer tokenseconds, lower is better 0.77 22 of 137 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: τ-bench banking% of tasks, higher is better 14.8% 105 of 175 Gemma 4 31B 20 Oct 2026
Artificial Analysis: τ-bench banking% of tasks, higher is better 8.9% 133 of 175 Gemma 4 31B (no reasoning) 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 59.9% 151 of 286 Gemma 4 31B 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 65.5% 140 of 286 Gemma 4 31B (no reasoning) 20 Oct 2026

Not shown: Not your codebase or tools.

SimpleQA Verified (Epoch AI) · reported by SimpleQA Verified (Epoch AI) · 1 result
MeasureValueRankConfigurationDated
SimpleQA Verified: correct answers% of questions, higher is better 10.4% 83 of 85 Gemma 4 31B 27 Aug 2026

Not shown: Not answers grounded in your documents; tests what the model remembers.

UGI Leaderboard · reported by UGI Leaderboard · 6 results
MeasureValueRankConfigurationDated
UGI: requested-length error% off the requested word count, lower is better 5.0% 41 of 379 Gemma 4 31B 12 Apr 2026
UGI: requested-length error% off the requested word count, lower is better 6.0% 49 of 379 Gemma 4 31B (channel-thought-prefill reasoning) 12 Apr 2026
UGI: style adherencescore from 0 to 1, higher is better 0.34 218 of 379 Gemma 4 31B 12 Apr 2026
UGI: style adherencescore from 0 to 1, higher is better 0.36 145 of 379 Gemma 4 31B (channel-thought-prefill reasoning) 12 Apr 2026
UGI: writing scorescore out of 100, higher is better 38.6 234 of 379 Gemma 4 31B 12 Apr 2026
UGI: writing scorescore out of 100, higher is better 43.1 199 of 379 Gemma 4 31B (channel-thought-prefill reasoning) 12 Apr 2026

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

Vectara Hallucination Leaderboard · reported by Vectara Hallucination Leaderboard · 2 results
MeasureValueRankConfigurationDated
Vectara: answer rate% of documents, higher is better 100.0% 1 of 108 Gemma 4 31B 22 Sep 2026
Vectara: hallucination rate% of summaries, lower is better 7.4% 35 of 108 Gemma 4 31B 22 Sep 2026

Not shown: Not a quality score: a low rate usually means content filters were triggered, and hallucination rates are measured on answered documents only.

Will Gemma 4 31B work for your product?

Benchmarks aren't your task. Paste the prompt your product uses and run it beside the models you could use instead, with every answer checked and costed.

New here? Your email starts a free trial. No credit card needed.