Gemma 4 31B
GoogleReleased Apr 2, 2026Open weights · Apache 2.0gemma-4-31b
- Intelligence (Epoch)
- 143 · 71st of 111
- Reasoning
- yes
- Knowledge cutoff
- January 2025
- Context
- 262K tokens
- Max output
- 128k tokens
- Input
- text, image, video, audio
- API features
- tools, structured outputs, JSON mode, seed, logprobs
Read Oct 23, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from Google · cutoff: ai.google.dev
Free right now
- Free OpenRouter, served by Google AI Studio · no end date · 20 req/min, 50 req/day, 1,000 req/day with $10 of credits
Checked Oct 23 · every free model on Deals
The cheapest host, Reka, serves a weights precision we can't read (host terms). Cheapest at the maker's precision and context: Crusoe, $0.54 blended. Weights and context per host
Intelligence for the price
5 models score higher for the same price or less; it ranks 6 of the 20 models at its price or below.
Epoch Capabilities Index against the lowest input + output price per M, log scale · prices Oct 23. Score: Epoch AI, CC BY 4.0, 143 with an interval of 140 to 146.
Against its rivals
The best of its price range and the top score
- Gemma 4 31Bthis model$0.38143
- Claude Opus 5.5top score$24.00169
- Claude Haiku 5.5same price range$0.60159
- GPT-6 Lunasame price range$0.60157
Same level for 2.4× less: Qwen3.7 Flash at Alibaba, $0.16 per M input + output · their Epoch intervals overlap (144 against 143).
Bars: Epoch Capabilities Index (Epoch AI, CC BY 4.0), drawn from 100, below the lowest score in the catalogue · price: lowest input + output per M.
Speed by provider
Output speed of Gemma 4 31B at each host, fastest first
- SambaNova168 tok/s1.75 s to first token
- Parasail53 tok/s0.77 s to first token
- Crusoe51 tok/s0.46 s to first token
- Friendli49 tok/s3.96 s to first token
- CoreWeave46 tok/s0.54 s to first token
- Venice39 tok/s0.73 s to first token
- Io Net34 tok/s0.76 s to first token
- DeepInfra25 tok/s0.41 s to first token
- Chutes17 tok/s4.08 s to first token
- SiliconFlow8 tok/s1.29 s to first token
Speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · a snapshot of real traffic before the daily read, not a daily average · bars from zero.
Run it yourself
open weights, Apache 2.0 · fits 6 of the 6 workstations in our guideSelf-host guide ›- Parameters
- 31B
- dense
- Memory needed
- 19 GB
- in INT4 · 37 GB in INT8
- Smallest cloud setup
- RTX 3090
- 24 GB · $0.17/h, about $124 a month
- Worth it above
- 327M tokens
- a month, against the cheapest API at $0.38 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 19 GB | RTX 3090 · $0.17/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 37 GB | A100 · $1.19/h | near-lossless |
| FP16 (full) | 74 GB | A100 · $1.19/h | reference quality |
At home: in INT4 it fits a RTX 5090 (32 GB) + host PC (plus a host PC), about $3,500 once, or $0.1332/h over 3 years, electricity aside; it pays for itself after about 9B tokens against the cheapest API.
Memory ≈ parameters × bytes per parameter × 1.2 · cloud price: on-demand rate, running 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| Chutesfp4 (bf16) · 128K ctx (256K) · 64K out | $0.12 | $0.37 | $0.012 | 17 | 4.08 s | 98.9% | Oct 23 |
| Rekaprecision not read: host terms · 256K ctx · 32K out | $0.08 | $0.30 | $0.05 | — | — | 99.8% | Sep 29 |
| DekaLLM256K ctx | $0.10 | $0.33 | $0.05 | — | — | 97.4% | Oct 4 |
| Parasailfp8 (bf16) · 256K ctx | $0.15 | $0.40 | $0.06 | 53 | 0.77 s | 99.9% | Oct 23 |
| Venicefp4 (bf16) · 250K ctx · 8K out | $0.12 | $0.36 | $0.09 | 39 | 0.73 s | 100% | Oct 23 |
| CoreWeavefp4 (bf16) · 256K ctx | $0.10 | $0.34 | $0.10 | 46 | 0.54 s | 100% | Oct 23 |
| Crusoebf16 · 256K ctx | $0.14 | $0.40 | $0.14 | 51 | 0.46 s | 99.6% | Oct 23 |
| Io Netprecision not read: host terms · 256K ctx · 16K out−5% | $0.36 | $1.09 | $0.18 | 34 | 0.76 s | 98.7% | Oct 23 |
| SiliconFlowfp8 (bf16) · 256K ctx | $0.75 | $1.00 | $0.25 | 8 | 1.29 s | 99.1% | Oct 23 |
| Friendliprecision not read: host terms · 256K ctx · 8K out | $0.14 | $0.40 | — | 49 | 3.96 s | 100% | Oct 23 |
| Novita AIbf16 · 256K ctx · 128K out | $0.14 | $0.40 | — | — | — | — | Oct 23 |
| DeepInfrafp8 (bf16) · 256K ctx · 16K out | $0.20 | $0.40 | — | 25 | 0.41 s | 99.2% | Oct 23 |
| Together AIfp8 (bf16) · 256K ctx | $0.39 | $0.97 | — | — | — | — | Sep 22 |
| SambaNovaprecision not stated · 128K ctx (256K) | $0.38 | $1.15 | — | 168 | 1.75 s | 96.0% | Oct 23 |
10 of 14 providers · prices per million tokens, read Oct 23, 06:00 UTC · speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · uptime: OpenRouter, last 24 h · −N%: a discount off the host's list price · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| Reka | not read: host terms | 256K | 32K | — | — | — |
| DekaLLM | not checked yet | 256K | not checked yet | — | — | — |
| CoreWeave | fp4 | 256K | not checked yet | yes | yes | yes |
| Venice | fp4 | 250K | 8K | yes | yes | yes |
| Chutes | fp4 | 128K | 64K | yes | yes | yes |
| Crusoe | bf16 | 256K | not stated by the host | yes | yes | yes |
| Friendli | not read: host terms | 256K | 8K | yes | yes | yes |
| Novita AI | bf16 | 256K | 128K | yes | yes | yes |
| Parasail | fp8 | 256K | not read: host terms | yes | yes | yes |
| DeepInfra | fp8 | 256K | 16K | yes | yes | yes |
| Together AI | fp8 | 256K | not stated by the host | — | — | — |
| Io Net | not read: host terms | 256K | 16K | yes | no | yes |
| SambaNova | not stated by the host | 128K | not stated by the host | yes | no | yes |
| SiliconFlow | fp8 | 256K | not read: host terms | yes | yes | yes |
Weights, context and max output: read at each host's own catalogue or docs where its terms allow, otherwise as the host declares to OpenRouter; hover a cell for its source and date · tools, structured outputs, reasoning: as declared to OpenRouter · not verified by us
Gemma 4 31B spec sheet, field by field, with sources
Gemma 4 31B spec sheetmaker-declared
What Google declares about the model itself — read from its own weights repository and documentation, not from any host. What each endpoint actually serves is in the pricing above: the two can differ, and both are true.
- Context window
- 262,144 tokens
- Parameters
- 31,273,088,876≈ 31.3B
- Active experts
- none declared in the weights
- Weight precision
- BF16
- License
- Apache 2.0 — open weights, permissivecurated · read 2026-09-07Scope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- January 2025maker's docs
- Modalities
- text · image · video · audio
- Max output (official)
- to verify
- Release date
- to verify≈ April 2, 2026 third-party catalogue — not the maker's figure
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- Gemma4ForConditionalGeneration
- Layers
- 60
- Vocabulary
- 262,144 tokens
- Weights first published
- March 11, 2026
- Weights last updated
- July 20, 2026
- Paper
- arXiv:2607.02770
google/gemma-4-31b-it · config.json @ 842da37 · read 2026-10-08 · 6 of 10 fields published by the maker
| Date | Provider | Input | Output | Change |
|---|---|---|---|---|
| Oct 22 | DeepInfra | $0.14 → $0.20 | $0.51 → $0.40 | −7% |
| Oct 16 | DeepInfra | $0.09 → $0.14 | $0.34 → $0.51 | +50% |
| Oct 6 | Io Net | $0.38 → $0.36 | $1.15 → $1.09 | −5% |
| Sep 14 | SiliconFlow | $0.13 → $0.75 | $0.40 → $1.00 | +230% |
| Sep 5 | DeepInfra | $0.27 → $0.09 | $0.76 → $0.34 | −58% |
Every published change of input or output price, newest first · change on input + output · full series: pricing-history.csv.
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| Reka | $0.23 | $0.79 | $0.62 |
| DekaLLM | $0.27 | $0.97 | $0.73 |
| CoreWeave | $0.27 | $0.97 | $0.74 |
| Venice | $0.30 | $1.14 | $0.84 |
| Chutes | $0.30 | $1.15 | $0.85 |
| Crusoe | $0.34 | $1.32 | $0.96 |
| Friendli | $0.34 | $1.32 | $0.96 |
| Novita AI | $0.34 | $1.32 | $0.96 |
| Parasail | $0.35 | $1.40 | $1.00 |
| DeepInfra | $0.40 | $1.80 | $1.20 |
| Together AI | $0.88 | $3.60 | $2.53 |
| Io Net | $0.91 | $3.43 | $2.54 |
| SambaNova | $0.95 | $3.62 | $2.67 |
| SiliconFlow | $1.25 | $6.50 | $4.00 |
What each host bills for 1,000 requests of a representative job
Your own workload
Also serving Gemma 4 31B, without a published price
No price we can verify, so no rank.
Other Google models · all its models, hosts and history ›
- Gemma 4 26B A4B12 providers · from $0.26
- Gemma 3 27B5 providers · from $0.24
- Gemma 3 12B3 providers · from $0.15
- Gemma 3 4B2 providers · from $0.09
- Gemini 2.5 Flash1 provider · from $2.80
- Gemini 2.5 Pro1 provider · from $11.25