Gemma 4 26B A4B (free)
Sold by 15 sellers, timed at 76.5 tok/s, scored 39.3 on the Coding Index, and with results on 1 openly licensed board.
retrieved sha256 · 0bb53a77713f
What sellers charge
15 sellers · 3x from cheapest to dearestOne tick per seller, on a log scale of input price per million from $0.05 at Darkbloom to $0.15 at Google. 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 bf16 group alone spans 70% 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
562,169 requests · 15 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
15 priced · 15 timed · cheapest first- Darkbloomdarkbloom/int4$0.05$0.19$0.0125int4131k52 tok/s30672ms122,404 req99.93%
- DekaLLMdekallm/bf16$0.06$0.33$0.04bf16236k44 tok/s271.02s3,773 req97.70%
- Io Netio-net/bf16$0.067$0.29$0.039bf16236k71 tok/s29809ms9,872 req100%
- NextBitnextbit/bf16 · 25% off$0.0675$0.225$0.0375bf16236k71 tok/s43449ms148,346 req100%
- AkashMLakashml/fp4$0.07$0.24$0.046fp4118k101 tok/s51470ms12,244 req100%
- DeepInfradeepinfra/fp8$0.07$0.34not listedfp816k53 tok/s30308ms26,939 req100%
- Makoramakora$0.09$0.28$0.049not declared128k106 tok/s54518ms86,274 req100%
- Cloudflarecloudflare$0.10$0.30$0.05not declared230k45 tok/s32848ms8,146 req95.62%
- CoreWeavecoreweave/bf16$0.10$0.30$0.05bf16236k102 tok/s58248ms86,463 req99.30%
- PrimeIntellectprimeintellect/bf16$0.10$0.30$0.05bf1633k56 tok/s12523ms7,488 req100%
- Novitanovita/bf16$0.13$0.40not listedbf16131k79 tok/s31831ms3,660 req100%
- Parasailparasail/bf16$0.13$0.40$0.05bf16236k84 tok/s51537ms31,478 req100%
- Venicevenice/bf16$0.13$0.40$0.05bf1616k88 tok/s45.9559ms4,728 req99.92%
- SiliconFlowsiliconflow/fp8$0.14$0.40$0.05fp8236k63 tok/s291.43s1,806 req100%
- Googlegoogle-vertex/global$0.15$0.60not listednot declared33k43 tok/s271.36s8,548 req96.07%
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: 9 of the 15 timed routes are slower than it, and the middle route runs at 71 tok/s. The widest median-to-p90 gap is PrimeIntellect's, 4.67x.
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 . 9 of the 16 sellers with a reading fell under 99% on at least one of the 23 days held, 23 Oct 2026 to 21 Nov 2026: Novita at 51.59% on 6 Nov 2026, Google at 71.57% on 19 Nov 2026, CoreWeave at 92.01% on 16 Nov 2026, SiliconFlow at 92.43% on 2 Nov 2026, and 5 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 · 15 sellers- include_reasoningevery seller
- max_tokensevery seller
- reasoningevery seller
- temperatureevery seller
- top_pevery seller
- frequency_penalty14 of 15not PrimeIntellect
- presence_penalty14 of 15not SiliconFlow
- response_format14 of 15not Io Net
- stop14 of 15not SiliconFlow
- top_k13 of 15not DekaLLM and NextBit
- logprobs12 of 15not DeepInfra, PrimeIntellect, and SiliconFlow
- repetition_penalty12 of 15not DekaLLM, SiliconFlow, and Venice
- seed12 of 15not PrimeIntellect, SiliconFlow, and Venice
- tool_choice12 of 15not CoreWeave, Parasail, and SiliconFlow
- tools12 of 15not CoreWeave, Parasail, and SiliconFlow
- top_logprobs12 of 15not DeepInfra, PrimeIntellect, and SiliconFlow
- structured_outputs11 of 15
- logit_bias4 of 15
- min_p3 of 15only Cloudflare, DeepInfra, and PrimeIntellect
- max_completion_tokens1 of 15only PrimeIntellect
5 of these 20 are accepted by every one of the 15 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 117 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 · 2 results| Task | All runs | Score |
|---|---|---|
| GPQA diamond | 342 | 0.73 |
| OTIS Mock AIME 2024-2025 | 319 | 0.82 |
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 this run gemma-4-26b-a4b-it_minimal. 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 26B A4B (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.
