Gemma 4 31B vs Muse Ember
Full profiles: Gemma 4 31B (Google) and Muse Ember (Meta): every benchmark, price and rank.
Updated · Data from 9 sources
As of 16 Nov 2026, Gemma 4 31B and Muse Ember are neck and neck on quality (51.9 vs 52.1 on the BenchLeader Index, inside the ±8.7 margin), and Muse Ember is 1.3× cheaper at $0.200 vs $0.250 per 1M tokens blended. Choose Gemma 4 31B for reasoning (61 vs 51) and fast first answers (0.5 s vs 5.0 s); choose Muse Ember for knowledge (49 vs 24) and coding (60 vs 48), where it also streams 3×+ faster.
Key facts (as of 16 Nov 2026)
- Gemma 4 31B scores 51.9 on the BenchLeader Index and Muse Ember scores 52.1 (BenchLeader, 16 Nov 2026).
- Gemma 4 31B costs $0.250 per 1M tokens blended (list price, 16 Nov 2026).
- Muse Ember costs $0.200 per 1M tokens blended (list price, 16 Nov 2026).
- Muse Ember streams 3×+ faster than Gemma 4 31B (Artificial Analysis, 7-day median, 16 Nov 2026).
- Of the 1 benchmarks both models have, Muse Ember leads 1 (BenchLeader, 16 Nov 2026).
- On Arena WebDev, Gemma 4 31B is rated 1370 and Muse Ember 1609 (LMArena, as published by 16 Nov 2026).
Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | Muse Ember | 60 vs 48 category score; Arena WebDev 1609 vs 1370 | High |
| Reasoning | Gemma 4 31B | 61 vs 51 category score | High |
| Knowledge | Muse Ember | 49 vs 24 category score | High |
| Chat quality (human preference) | – | Only Gemma 4 31B has an Arena rating (54 category score) | Not comparable |
| Lowest cost | Muse Ember | $0.200 vs $0.250 per 1M blended; agentic coding session $0.008 vs $0.014 | – |
| Fastest first answer | Gemma 4 31B | 0.5 s (Gemma 4 31B) vs 5.0 s (Muse Ember (max)) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- Gemma 4 31B and Muse Ember (max) are neck and neck on quality (51.9 vs 52.1, within the ±8.7 margin).
- Gemma 4 31B is stronger in reasoning.
- Muse Ember (max) is stronger in coding, knowledge.
- Muse Ember (max) is 1.3× cheaper ($0.200 vs $0.250 per 1M blended).
- Muse Ember (max) streams 6.8× faster (437 vs 64 tokens per second).
At a glance
| Spec | Gemma 4 31B | Muse Ember |
|---|---|---|
| Developer | Meta | |
| Released | 2 Apr 2026 | 9 Nov 2026 |
| Context window (tokens) | 262,144 | 1,048,576 |
| Compared at | Gemma 4 31B | Muse Ember (max) |
| Open weights | Yes | No |
| Price in / out / cached input per 1M | $0.200 / $0.400 / $0.250 | $0.100 / $0.500 / $0.0040 |
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | Gemma 4 31B | Muse Ember (max) | Leader |
|---|---|---|---|
| Reasoning | 61.0 | 51.1 | Gemma 4 31B |
| Coding | 48.2 | 60.0 | Muse Ember |
| Agents & tools | – | 52.4 | not comparable |
| Maths | 64.7 | – | not comparable |
| Knowledge | 24.4 | 48.8 | Muse Ember |
| Human preference | 54.3 | – | not comparable |
| Multimodal | 62.0 | – | not comparable |
On one chart
Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.
Not drawn, measured for fewer than two of these models: Agents & tools, Maths, Instruction following, Long context, Multimodal, Human preference.
Where sources disagree
Across the 1 benchmarks both have, Gemma 4 31B leads 0, Muse Ember leads 1. Every overall index that rates both puts Muse Ember ahead.
| Overall index | Gemma 4 31B | Muse Ember (max) | Leader |
|---|---|---|---|
| BenchLeader Index | 51.9 | 52.1 | neck and neck |
| AA Intelligence Index v5.0.1 | 4.4† | 36.0 | Muse Ember |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.
| Measure | Gemma 4 31B | Muse Ember (max) |
|---|---|---|
| Blended price per 1M tokens | $0.250 | $0.200 |
| Input / output per 1M | $0.200 / $0.400 | $0.100 / $0.500 |
| Cached input per 1M | $0.250 | $0.0040 |
| Price per task: chat reply | $0.0002 | $0.0002 |
| Price per task: summarize a 30-page report | $0.0026 | $0.0015 |
| Price per task: code edit | $0.0018 | $0.0014 |
| Price per task: agentic coding session | $0.014 | $0.0080 |
| Price per task: structured extraction | $0.0005 | $0.0003 |
| Agentic coding session, input mostly cached | $0.016 | $0.0037 |
| Cost per benchmark task (AA Index) | – | $0.136 |
| Output tokens per task (AA Index) | – | 84k |
| Of which reasoning | – | 52k |
| Time per task (AA Index) | – | 3 min |
| Output speed | 64 tok/s | 437 tok/s |
| Time to first answer | 0.5 s | 5.0 s |
| Time to a complete answer, last measured 13 Nov 2026 | – | 12.9 s |
Each model is compared at its best-ranked reasoning-effort setting. Gemma 4 31B scores 51.9 and takes 1 s to its first answer token; Gemma 4 31B (no reasoning) scores 49.8 in 4 s, Gemma 4 31B (thinking) scores 46.6 in 52 s. Muse Ember (max) scores 52.1 and takes 5 s to its first answer token; Muse Ember (xhigh) scores 49.1 in 4 s. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
94 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | Gemma 4 31B | Muse Ember (max) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 51.9 | 52.1 |
| Coding score | BenchLeader | 48.2 | 60.0 |
| Human preference score | BenchLeader | 54.3 | – |
| Knowledge score | BenchLeader | 24.4 | 48.8 |
| Maths score | BenchLeader | 64.7 | – |
| Multimodal score | BenchLeader | 62.0 | – |
| Reasoning score | BenchLeader | 61.0 | 51.1 |
| Agents & tools score | BenchLeader | – | 52.4 |
| Blended price $/M | BenchLeader | $0.250 | $0.200 |
| Output speed | BenchLeader | 64 tok/s | 437 tok/s |
| Time to first answer | BenchLeader | 0.5 s | 5.0 s |
| Context window | BenchLeader | 262,144 tokens | 1,048,576 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 75.8%† | – |
| SimpleBench | SimpleBench | – | 72.7%† |
| Arena Hard Prompts | Arena (LMArena) | 1475 | – |
| LiveBench 2.0 Reasoning | LiveBench | 68.3%† | – |
| GPQA Diamond (AA) | Artificial Analysis | 76.3%† | – |
| Humanity's Last Exam (AA) | Artificial Analysis | 11.8%† | – |
| HLE-Diamond (AA) | Artificial Analysis | 7.2%† | 25.1% |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 63.5% | – |
| CritPt | Artificial Analysis | 0.0%† | – |
| CritPt v2 hard subset (AA) | Artificial Analysis | 4.0%† | 39.0% |
| Arena Expert | Arena (LMArena) | 1498 | – |
| LMCA | Epoch AI Benchmarking Hub | 39.3% | – |
| DTBench | Epoch AI Benchmarking Hub | 82.7% | – |
| ForecastBench | ForecastBench | 61.1% | – |
| Coding | |||
| SciCode | SciCode | 45.5% | – |
| WeirdML | WeirdML | 52.3% | – |
| Arena Coding | Arena (LMArena) | 1502 | – |
| Arena WebDev | Arena (LMArena) | 1370 | 1609 |
| LiveBench 2.0 Coding | LiveBench | 20.3%† | – |
| Vibe Code Bench v1.1 | Vals AI | – | 83.7% |
| SWE Atlas v1.1 | Scale AI SEAL | – | 56.0%† |
| SWE Atlas v1.1: Codebase QnA | Scale AI SEAL | – | 61.3%† |
| SWE Atlas v1.1: Test Writing | Scale AI SEAL | – | 56.7%† |
| SWE Atlas v1.1: Refactoring | Scale AI SEAL | – | 50.0%† |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | 30.6% | – |
Show all 94 benchmarks
| Metric | Publisher | Gemma 4 31B | Muse Ember (max) |
|---|---|---|---|
| Coding | |||
| ALE-Bench | Epoch AI Benchmarking Hub | 925.5 | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 0.0%† | 40.9% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 29.2%† | – |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 18.6† | – |
| Design Arena Website | Design Arena | – | 1330† |
| FrontierCode 2.0 (AA) | Artificial Analysis | 0.0%† | 27.2% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | – | 30.6%† |
| APEX-Agents | Mercor | – | 64.3%† |
| Arena Agent | Arena (LMArena) | – | 1.7 |
| LiveBench Long Agentic | LiveBench | 8.5%† | – |
| LiveBench Browser | LiveBench | 0.0%† | – |
| Terminal-Bench Hard | Artificial Analysis | 30.3%† | – |
| τ²-Bench Telecom (AA) | Artificial Analysis | 65.5%† | – |
| GDPval-AA v2.1, normalised score | Artificial Analysis | 2.4%† | – |
| GDPval-AA v2.1 | Artificial Analysis | 555† | – |
| τ³-Banking (AA) | Artificial Analysis | 8.9%† | – |
| ITBench SRE (AA) | Artificial Analysis | 37.3%† | – |
| Finance Agent v2.1 | Vals AI | – | 63.8% |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | – | 25.0% |
| Legal Research Bench | Vals AI | – | 49.0% |
| Tax Agent Bench | Vals AI | – | 38.2% |
| Terminal-Bench 2.0 (Vals) | Vals AI | 39.3%† | – |
| Terminal-Bench 4.0 (Vals) | Vals AI | – | 39.4% |
| Arena Agent Task Completion | Arena (LMArena) | – | -7.6 |
| AutomationBench | Artificial Analysis | 4.9%† | 68.1% |
| GDP.pdf (AA) | Artificial Analysis | 6.0%† | 22.8% |
| EnterpriseOps-Gym | Artificial Analysis | 28.4%† | – |
| AA-Briefcase v1.1 | Artificial Analysis | 369† | 1520 |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 0.0%† | 20.0% |
| FACTS Search | 75.3% | – | |
| Maths | |||
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 73.3%† | – |
| Arena Maths | Arena (LMArena) | 1467 | – |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 10.4% | – |
| LiveBench 2.0 Data Analysis | LiveBench | 53.3%† | – |
| LiveBench Research | LiveBench | 38.1%† | – |
| AA-Omniscience v1.1 | Artificial Analysis | -48.5† | 7.4 |
| CaseLaw v2 | Vals AI | 52.6%† | – |
| Excel Modeling Benchmark v1.1 | Vals AI | – | 72.2% |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | 19.9%† | 21.1% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | 14.5%† | 82.6% |
| FACTS Parametric | 6.9% | – | |
| Instruction following | |||
| IFBench | Artificial Analysis | 53.5%† | – |
| Arena Instruction Following | Arena (LMArena) | 1453 | – |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1452 | – |
| EQ-Bench 4 | EQ-Bench | 1110 | – |
| EQ-Bench Creative Writing v3 | EQ-Bench | 1369 | – |
| Arena Creative Writing | Arena (LMArena) | 1418 | – |
| Arena Multi-turn | Arena (LMArena) | 1465 | – |
| Arena Longer Queries | Arena (LMArena) | 1469 | – |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | 1279 | – |
| LiveBench Design | LiveBench | 56.5%† | – |
| MMMU-Pro | Artificial Analysis | 70.3%† | – |
| MortgageTax | Vals AI | 61.4%† | – |
| SAGE | Vals AI | 82.8%† | – |
| FACTS Multimodal | 42.3% | – | |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 46.7%† | – |
| Arena Document | Arena (LMArena) | 1452 | – |
| MLCR | Artificial Analysis | – | 18.3% |
| FACTS Grounding | 75.8% | – | |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 144.3 | – |
| LiveBench 2.0 | LiveBench | 32.4%† | – |
| AA Intelligence Index v5.0.1 | Artificial Analysis | 4.4† | 36.0 |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 13.9† | – |
| AA Progress Index v1 | Artificial Analysis | 233† | – |
| Vals Index | Vals AI | – | 60.5 |
| Vals Finance Index | Vals AI | – | 60.9% |
| Vals Legal Index | Vals AI | – | 75.7% |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | – | 11.9%† |
| Metadata | |||
| AA Openness Index | Artificial Analysis | 38.9† | – |
Data as of 2026-11-16. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261111-7b598bb13de4 (https://benchmarks.duellab.org/releases/gb2-20261111-7b598bb13de4/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- Gemma 4 31B: BenchLeader Index 47.5 → 51.9.
- Gemma 4 31B: 1 new result (FACTS Search).
- Muse Ember (max): 1 new result (Arena Agent).
- Muse Ember (max): 2 new results (MLCR, Arena WebDev).
- Gemma 4 31B: BenchLeader Index 48.0 → 47.5.
- Gemma 4 31B: 5 score revisions.
- Gemma 4 31B: 1 score revision.
- Gemma 4 31B: 1 score revision.
- Gemma 4 31B: BenchLeader Index 48.5 → 48.0.
- Gemma 4 31B: 3 score revisions.
Related comparisons
- Muse Ember vs Muse Spark Open 816B A42B
- Muse Ember vs Mistral Large 4
- Muse Ember vs Gemini 3.1 Flash Lite
- Muse Ember vs Inkling Small
- Muse Ember vs Qwen3.5-27B
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
Gemma 4 31B vs Muse Ember: questions
- Is Gemma 4 31B better than Muse Ember?
- Gemma 4 31B and Muse Ember (max) are neck and neck on quality (51.9 vs 52.1, within the ±8.7 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 16 Nov 2026, so check the category scores for your use.
- Gemma 4 31B vs Muse Ember: which is cheaper, and will that change?
- Muse Ember is cheaper today: $0.200 against $0.250 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Gemma 4 31B or Muse Ember better for coding?
- Muse Ember scores higher in coding (60 vs 48 on the category index, where 50 is average). On Arena WebDev, the top shared coding benchmark, it is 1370 for Gemma 4 31B against 1609 for Muse Ember.
- Which is faster, Gemma 4 31B or Muse Ember?
- Muse Ember streams 3×+ faster (437 vs 64 output tokens per second, Artificial Analysis 7-day median).
- Which is better for long documents?
- Muse Ember accepts more context: 1,048,576 against 262,144 tokens for Gemma 4 31B.
More questions (3)
- Which has lower latency?
- Gemma 4 31B answers first: 0.5 s against 5.0 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Do independent benchmarks agree on Gemma 4 31B vs Muse Ember?
- Across the 1 benchmarks both have, Gemma 4 31B leads 0, Muse Ember leads 1. Every overall index that rates both puts Muse Ember ahead.
- When was Muse Ember released?
- Muse Ember was released on 9 Nov 2026.