Gemma 4 31B (free)
Sold by 18 sellers, timed at 87.5 tok/s, scored 4.4 on the Intelligence Index, and with results on 1 openly licensed board.
retrieved sha256 · 0bb53a77713f
What sellers charge
18 sellers · 7.5x from cheapest to dearestOne tick per seller, on a log scale of input price per million from $0.10 at CoreWeave to $0.75 at SiliconFlow. A faint tick declares no number format. Every seller is named on the endpoint ledger below.
The same sellers grouped by the number format each declares, on the same scale. The fp4 group alone spans 100% of the range, so the cheap end is not cheap for its number format. A declaration is checked by nobody: this groups the rows and certifies none.
Measured speed
292,329 requests · 18 endpoints · 30-minute windowOpenRouter's own timing of its routed traffic, weighted by requests, over one 30-minute window read . Every route is on the endpoint ledger below.
Every endpoint
18 priced · 18 timed · cheapest first- CoreWeavecoreweave/fp4$0.10$0.34$0.10fp4236k63 tok/s37417ms62,408 req100%
- KTkt$0.10$0.34$0.10not declared236k129 tok/s51694ms51,772 req100%
- PrimeIntellectprimeintellect/fp8$0.105$0.177$0.053fp816k34 tok/s21407ms1,572 req99.64%
- Chuteschutes/fp4$0.12$0.37$0.012fp466k17 tok/s122.91s2,685 req99.64%
- Venicevenice/fp4$0.12$0.36$0.09fp416k49 tok/s30707ms5,163 req99.93%
- NextBitnextbit/fp4$0.125$0.34$0.075fp4236k82 tok/s26472ms34,431 req100%
- DeepInfradeepinfra/turbo$0.135$0.51$0.075fp416knot matchednonenonenot matched98.88%
- Crusoecrusoe/bf16$0.14$0.40$0.14bf16262k47 tok/s23615ms4,216 req100%
- Friendlifriendli$0.14$0.40not listednot declared8k140 tok/s771.52s11,423 req100%
- Novitanovita/bf16$0.14$0.40not listedbf16131k79 tok/s51629ms6,951 req100%
- Parasailparasail/fp8$0.15$0.40$0.06fp8236k47 tok/s25595ms16,339 req100%
- DeepInfradeepinfra/fp8/flex$0.16$0.32not listedfp8236knot matchednonenonenot matched100%
- DeepInfradeepinfra/fp8$0.20$0.40not listedfp816knot matchednonenonenot matched99.85%
- DeepInfradeepinfra/fp8/priority$0.30$0.60not listedfp8236knot matchednonenonenot matched99.02%
- Modalmodal$0.34$0.10$0.10not declared118k112 tok/s46396ms3,121 req100%
- SambaNovasambanova$0.38$1.15not listednot declared118k186 tok/s1011.4s364 req93.54%
- ModelRunmodelrun/fp4$0.75$1.00$0.20fp4236k219 tok/s151145ms22,859 req100%
- SiliconFlowsiliconflow/fp8$0.75$1.00$0.25fp8236k50 tok/s431.06s648 req100%
- DeepInfranot matchednot matchednot matchednot matchedno offer53 tok/s24389ms56,099 reqno offer
- DeepInfranot matchednot matchednot matchednot matchedno offer21 tok/s12693ms11,738 reqno offer
- DeepInfranot matchednot matchednot matchednot matchedno offer25 tok/s161.56s39 reqno offer
- DeepInfranot matchednot matchednot matchednot matchedno offer21 tok/s12632ms501 reqno offer
Each track runs from a route's p90, the rate nine in ten of its requests met or beat, up to its median. Weighted by requests, the figure above reports where the traffic went rather than what one route does: 13 of the 18 timed routes are slower than it, and the middle route runs at 51.5 tok/s. The widest median-to-p90 gap is NextBit's, 3.15x.
DeepInfra publishes 4 offers and had 4 routes timed. The two documents share no route id, so those offers and routes are printed apart rather than paired.
Uptime is OpenRouter's own figure for each seller over the trailing day, read once a day, and the column prints the latest reading, taken . 17 of the 21 sellers with a reading fell under 99% on at least one of the 23 days held, 23 Oct 2026 to 21 Nov 2026: SambaNova at 64.75% on 3 Nov 2026, Novita at 65.36% on 23 Oct 2026, DekaLLM at 77.08% on 13 Nov 2026, Modal at 85.63% on 16 Nov 2026, and 13 more. Source: OpenRouter's public endpoints API, one reading a day for each seller: 1,211 sellers carried one on the latest of the 23 days held.
Source: OpenRouter's public endpoints API. 1257 sellers across 359 models when the mirror ran, 186 of them served by more than one. List prices, not what anybody paid; a discount is the seller's own. Read . Speed is OpenRouter's own timing over one 30-minute window, read .
What each seller accepts
20 parameters · 18 sellers- include_reasoningevery seller
- max_tokensevery seller
- reasoningevery seller
- temperatureevery seller
- top_pevery seller
- response_format17 of 18not SambaNova
- top_k17 of 18not NextBit
- frequency_penalty16 of 18not Modal and SambaNova
- stop16 of 18not KT and SiliconFlow
- structured_outputs16 of 18not KT and SambaNova
- tool_choice16 of 18not DeepInfra (deepinfra/fp8/flex) and DeepInfra (deepinfra/fp8/priority)
- tools16 of 18not DeepInfra (deepinfra/fp8/flex) and DeepInfra (deepinfra/fp8/priority)
- presence_penalty15 of 18not Modal, SambaNova, and SiliconFlow
- repetition_penalty14 of 18
- seed12 of 18
- logprobs7 of 18
- min_p7 of 18
- top_logprobs7 of 18
- logit_bias6 of 18
- max_completion_tokens1 of 18only PrimeIntellect
5 of these 20 are accepted by every one of the 18 sellers that publish a list; the other 15 are declared by some sellers and not others, so what a request sending one of them gets depends on which seller serves it. Each list is the seller's own declaration in OpenRouter's endpoints API, read , and nothing here tested one.
Measured quality
145 models with an index · 132 on a design arena boardEach mark is one board this model stands on, placed at the share of that board it scores above, and led to its number in the list under the rule. The columns and the ledger below carry the figures.
Not ranked in any Design Arena category.
Position 102 of the 258 models Epoch has placed, which is not the population of either board above, on a scale anchored at Claude 3.5 Sonnet 130 and GPT-5 150. Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.
Benchmark results
1 of 6 boards · 3 results| Task | All runs | Effort | Score |
|---|---|---|---|
| GPQA diamond | 342 | minimal | 0.76 |
| OTIS Mock AIME 2024-2025 | 319 | minimal | 0.73 |
| SimpleQA Verified | 106 | unstated | 0.10 |
Mean accuracy, 0 to 1, on the task the row names, as Epoch AI runs it and averaged over repeated runs. All eighteen of their tasks, including the held-out FrontierMath tiers, named as they name them. The model id carries the effort setting the run used. Epoch AI Benchmarking Hub names these runs gemma-4-31b-it_minimal and gemma-4-31b-it, and the effort beside each score says which run it is, unstated where the publisher wrote none. A run missing from a task's rows has no published result there, which is not a score of zero. All runs is how many runs the publisher scored on the task, every model's. Read from epoch.ai ↗, which dates its own data . Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
1 of the 6 openly licensed boards Vitna mirrors carries a result for Gemma 4 31B (free), each printed as its publisher published it and none added to another. MTEB, LiveBench, HELM, Vectara hallucination leaderboard and Aider polyglot have not published one under a name that matches its listing, and names are matched to the catalog exactly and never guessed. A task that is a link opens its column on /capability, with every run scored on it. Every board, with its licence and what its publisher measures, is on /benchmarks.
Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Source: each model's own Hugging Face repository (huggingface.co), as of 20 Nov 2026. 162 repositories read, 4 did not answer.
Licence, parameter count, model class and gating: each model's own Hugging Face repository, the one OpenRouter names for the listing.
A licence is the tag the uploader put on the repository, read from nothing else and verified by nobody.
A parameter count is the total the repository's safetensors files declare; for a mixture of experts that is every expert, not the share one token runs through.
Source: the Epoch AI Capabilities Index (epoch.ai/eci), as of 20 Nov 2026. 258 models placed over 2917 benchmark results.
Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
Vitna ran none of these evaluations and adjusted none of these figures.
