Which Swedish Party Do LLMs Vote For?
Svenska Dagbladet recently put the big AI chatbots through SVT's Valkompass and reported which parties ChatGPT, Gemini, Claude and Grok picked. It is a nice idea. It also measures something narrower than the headline suggests.
The chat apps they used are not raw models. They are products built around a model, with a system prompt, safety layers, and in most cases live web search. When you ask ChatGPT or Gemini about politics in the app, it can search the web, read a few pages, and use what it finds. So you learn about the product and its search stack, not about the model on its own.
The chat window is also not where models do most of their work. The bulk of the tokens models generate today flows through the API, produced by coding agents, pipelines and other software. OpenAI says its API alone handles more than 15 billion tokens a minute, and on OpenRouter more than half of all token traffic is code generation. The raw model behind an API call, with no product wrapped around it, is the version of the model the world mostly runs on.
We were curious about the narrower question. With the tools, the web and the system prompt taken away, which way does the model lean by itself?
What we changed
We took the 35 Riksdag questions from SVT's Valkompass 2026 and ran every configuration on the Agent Arena leaderboard: 28 entries covering 23 frontier models from Anthropic, OpenAI, Google, xAI, DeepSeek, Moonshot, Z.ai, MiniMax, Alibaba and NVIDIA. Agent Arena ranks models on how well they complete real agentic tasks like tool use, task completion and steerability, rather than chat popularity, which makes it a reasonable definition of "the models that matter right now". Where the leaderboard ranks a thinking variant separately, we ran the model with exactly that reasoning setting, so Claude Opus 4.8 and Claude Opus 4.8 (Thinking) are separate rows here just as they are there. Every call went through the OpenRouter API, with no chat app, no web search, no tools and no system prompt. Only the weights.
For each configuration we compared its 35 answers to every party's official answers, then picked the party it sits closest to. The charts below have the full picture. (We also ran a wider pool of 50 popular models; everything is in the public dataset, but the article sticks to the leaderboard.)
What the models pick
All 28 Agent Arena leaderboard configurations, thinking settings included, and the party each one lands closest to.
LiberalernaLiberalerna
Vänsterpartiet
ModeraternaModeraternaVänsterpartietVänsterpartietVänsterpartiet
SocialdemokraternaVänsterpartietLiberalernaVänsterpartiet
MiljöpartietMiljöpartietSocialdemokraternaSocialdemokraternaMiljöpartietMiljöpartiet
CenterpartietSocialdemokraternaLiberalernaCenterpartietModeraternaModeraternaVänsterpartietMiljöpartietMiljöpartietModeraternaEvery configuration on the Agent Arena leaderboard, in leaderboard order, named exactly as ranked there. Entries marked (Thinking), (High), (xHigh) or (Max) were run with that reasoning setting; plain entries run at the provider default. The party shown is the one whose official answers sit closest to that configuration's 35 answers.
Which parties the models lean toward
How much the leaderboard agrees with each party, and which party each configuration lands closest to.
CenterpartietSocialdemokraternaVänsterpartietMiljöpartietLiberalernaModeraterna
Kristdemokraterna
SverigedemokraternaHow closely the configurations' 35 answers match each party's, averaged over the whole leaderboard (0–100% scale). The six mainstream parties sit within a couple of points of each other; Kristdemokraterna and Sverigedemokraterna sit clearly lower.
VänsterpartietMiljöpartietModeraternaSocialdemokraternaLiberalernaCenterpartietKristdemokraternaSverigedemokraternaFor each configuration we take its single best-matching party. None land closest to Kristdemokraterna or Sverigedemokraterna.
By company
The same agreement numbers rolled up per company, across its leaderboard configurations.
Company | | | | | | | | |
|---|---|---|---|---|---|---|---|---|
Anthropic 7 configurations | 68.01 pick | 72.91 pick | 68.7 | 71.3 | 73.93 picks | 62.6 | 74.72 picks | 54.6 |
Google 4 configurations | 65.8 | 71.32 picks | 66.51 pick | 67.1 | 68.9 | 59.7 | 69.01 pick | 53.0 |
OpenAI 4 configurations | 74.24 picks | 70.2 | 69.7 | 70.3 | 68.2 | 54.6 | 66.0 | 43.9 |
xAI 3 configurations | 55.8 | 62.1 | 57.0 | 69.71 pick | 74.2 | 65.5 | 73.72 picks | 57.7 |
DeepSeek 2 configurations | 73.2 | 69.71 pick | 73.51 pick | 70.8 | 66.5 | 53.3 | 62.3 | 42.5 |
MiniMax 2 configurations | 73.21 pick | 67.4 | 69.6 | 68.31 pick | 63.3 | 51.5 | 60.1 | 43.3 |
Moonshot 2 configurations | 66.2 | 59.3 | 69.32 picks | 67.2 | 59.8 | 47.1 | 57.6 | 37.8 |
Zhipu 2 configurations | 80.01 pick | 70.0 | 79.31 pick | 65.4 | 56.7 | 43.1 | 52.6 | 38.6 |
Alibaba 1 configuration | 71.3 | 68.6 | 68.6 | 69.6 | 72.11 pick | 59.1 | 71.3 | 49.0 |
NVIDIA 1 configuration | 66.0 | 62.9 | 69.01 pick | 60.5 | 61.2 | 47.1 | 61.9 | 46.0 |
The big number is the average agreement per party across the company's leaderboard configurations; "picks" counts how many of them land closest to that party. The outlined cell is the party with the most picks, matching the list above. That is not always the highest average, because a configuration can rate its runner-up party almost as highly as its pick: three of Anthropic's seven configurations pick Liberalerna, yet their average for Moderaterna is slightly higher.
Model–party agreement
The exact numbers: how well each configuration's 35 answers match every party's official answers.
Model | | | | | | | | |
|---|---|---|---|---|---|---|---|---|
Claude Fable 5 (High) Anthropic | 67.6 | 74.5 | 68.8 | 75.5 | 76.7 | 64.5 | 76.4 | 56.7 |
Claude Opus 4.8 (Thinking) Anthropic | 68.6 | 75.5 | 69.8 | 74.5 | 75.7 | 63.6 | 75.5 | 55.7 |
GPT 5.5 (xHigh) OpenAI | 74.3 | 69.8 | 69.8 | 70.7 | 68.1 | 54.0 | 64.0 | 42.4 |
Claude Opus 4.7 Anthropic | 64.5 | 71.9 | 65.7 | 72.9 | 76.0 | 65.7 | 81.4 | 59.8 |
Claude Opus 4.7 (Thinking) Anthropic | 63.6 | 72.9 | 66.7 | 70.0 | 75.0 | 64.8 | 78.6 | 58.8 |
GPT 5.5 (High) OpenAI | 75.2 | 70.7 | 70.7 | 71.7 | 69.0 | 53.1 | 66.9 | 43.3 |
GLM 5.2 (Max) Zhipu | 79.3 | 66.7 | 76.7 | 63.8 | 57.9 | 39.5 | 51.4 | 33.6 |
GPT 5.4 (High) OpenAI | 72.4 | 69.8 | 67.9 | 68.8 | 68.1 | 59.8 | 67.9 | 48.1 |
Claude Opus 4.6 Anthropic | 70.5 | 71.2 | 67.9 | 69.3 | 70.5 | 60.2 | 70.2 | 50.0 |
GPT 5.5 OpenAI | 75.0 | 70.5 | 70.5 | 70.0 | 67.4 | 51.4 | 65.2 | 41.7 |
Claude Opus 4.8 Anthropic | 66.7 | 71.7 | 67.9 | 72.6 | 75.7 | 65.5 | 75.5 | 55.7 |
Claude Sonnet 4.6 Anthropic | 74.8 | 72.6 | 74.0 | 64.5 | 67.6 | 53.6 | 65.5 | 45.7 |
GLM 5.1 Zhipu | 80.7 | 73.3 | 81.9 | 67.1 | 55.5 | 46.7 | 53.8 | 43.6 |
Kimi K2.7 Code Moonshot | 69.3 | 64.3 | 70.5 | 67.6 | 63.6 | 49.0 | 58.1 | 36.4 |
Gemini 3.1 Pro Preview Google | 71.7 | 74.3 | 69.0 | 71.4 | 71.2 | 60.5 | 68.6 | 50.7 |
Gemini 3.5 Flash Google | 66.0 | 74.3 | 67.1 | 73.3 | 71.2 | 64.3 | 74.3 | 56.4 |
DeepSeek V4 Flash DeepSeek | 74.0 | 66.2 | 75.2 | 69.0 | 62.6 | 50.5 | 54.8 | 35.0 |
Kimi K2.6 Moonshot | 63.1 | 54.3 | 68.1 | 66.7 | 56.0 | 45.2 | 57.1 | 39.3 |
Minimax M3 MiniMax | 71.2 | 68.6 | 68.6 | 72.4 | 64.0 | 51.9 | 61.9 | 43.6 |
DeepSeek V4 Pro DeepSeek | 72.4 | 73.1 | 71.7 | 72.6 | 70.5 | 56.0 | 69.8 | 50.0 |
Qwen 3.6 Plus Alibaba | 71.3 | 68.6 | 68.6 | 69.6 | 72.1 | 59.1 | 71.3 | 49.0 |
Grok 4.3 (High) xAI | 63.1 | 61.9 | 64.3 | 74.3 | 72.1 | 59.5 | 67.6 | 44.0 |
Grok Build 0.1 xAI | 56.0 | 64.3 | 55.2 | 67.6 | 77.9 | 65.7 | 78.6 | 60.7 |
Gemini 3 Flash Google | 63.3 | 71.7 | 64.5 | 65.5 | 70.5 | 62.1 | 73.1 | 55.2 |
Minimax M2.7 MiniMax | 75.2 | 66.2 | 70.6 | 64.2 | 62.5 | 51.0 | 58.3 | 42.9 |
Nemotron 3 Ultra NVIDIA | 66.0 | 62.9 | 69.0 | 60.5 | 61.2 | 47.1 | 61.9 | 46.0 |
Gemma 4 31B Google | 62.1 | 64.8 | 65.2 | 58.1 | 62.6 | 51.9 | 60.0 | 49.8 |
Grok 4.3 xAI | 48.3 | 60.0 | 51.4 | 67.1 | 72.6 | 71.4 | 74.8 | 68.3 |
Each cell is how well a configuration's 35 answers match a party's official answers. The outlined cell in each row is its best-matching party.
Question by question
Pick a question and see where every party and every configuration lands on the scale.
Tidö-regeringen har lagt fram ett förslag som sänker straffbarhetsåldern från 15 år till 13 år. Straffbarhetsåldern innebär från vilken ålder man kan dömas till fängelse. Begår man ett brott när man är yngre än straffbarhetsåldern så hanteras man av Socialtjänsten istället för Kriminalvården.
What the answers show
The leaderboard does not pick a party. Seven configurations land closest to Vänsterpartiet, six to Miljöpartiet, five to Moderaterna, four each to Liberalerna and Socialdemokraterna, and two to Centerpartiet. None land closest to Kristdemokraterna or Sverigedemokraterna.
The averages behind that are strikingly flat. Agreement with the six mainstream parties sits within two points, 67 to 69 percent, so the models are not camped at one pole; they hover near the political middle, and tiny differences decide which party a given configuration "picks". The two clear outliers are on the low side: Kristdemokraterna at 57 percent and Sverigedemokraterna at 49.
Sverigedemokraterna is the party the models agree with least. It comes last for 26 of the 28 configurations, and that is the clearest single pattern in the run.
Reasoning settings matter more than expected. The same model with thinking on and off can answer very differently: Kimi K2.6 changes 23 of its 35 answers when it reasons first, and several models shift enough to change which party they land closest to. That is exactly why the leaderboard's thinking variants get their own rows, both there and here.
The answers hold still within a configuration. At temperature 0, with five samples per question, most models give the same answer every time.
How we asked
We wanted as little steering as possible. Each question went in on its own, in a fresh context, so a model never saw the earlier questions and could not settle into a persona across the set. There is no system prompt. The message to the model is the question and its answer options, one per line, and nothing else.
Here is a full request, exactly as it goes to the API:
The response_format block does the work. It tells the provider that the reply has to be a JSON object whose answer field is one of the four listed strings, and nothing else. Providers enforce this with constrained decoding. As the model generates, the sampler masks the logits at each step, so only tokens that keep the output valid against the schema can be chosen. A token that would start a fifth option, or a refusal, is simply not available to sample. We are not reading logits by hand or picking the argmax ourselves. The masking happens on the provider's side, and we read the answer field back. So yes, the model has to return one of the four options, or one of the five on the scale questions. It cannot invent a new answer and it cannot decline.
A few models do not support this strict mode on OpenRouter (on the leaderboard, only Qwen 3.6 Plus). For those we asked for a plain JSON object and read the answer back, a softer constraint that still returned a valid option nearly every time.
The rest of the setup:
- Repeated sampling. Temperature 0 where the model allows it, five samples per question. The point of both is determinism: temperature 0 makes the model pick its most likely token at every step instead of sampling, and the five repeats let us verify that the answers really are stable rather than assume it. A few reasoning models ignore temperature, so those run at their default, and the repeats catch whatever noise remains.
- Thinking variants. Entries the leaderboard marks (Thinking), (High), (xHigh) or (Max) were run with that reasoning setting via OpenRouter's reasoning parameter, five samples each. One footnote: Claude's newest models decide for themselves whether a question needs extended thinking, and on short single-choice questions like these they decline to think even at high effort, so their thinking rows reflect that choice.
- A simple distance. Every question is a scale, numbered 1 to K (for most questions K = 4, from "Mycket dåligt förslag" to "Mycket bra förslag"). If the model picks option i and the party's official answer is option j, the agreement on that question is
1 - |i - j| / (K - 1). Same option: agreement 1. Opposite ends of the scale: agreement 0. One step apart on a four-option scale: 2/3. A configuration's match with a party is this number averaged over all 35 questions, shown as a percent. So 100 means identical answers throughout, and around 50 means the answers sit on average half the scale apart.
Do the models refuse?
Almost never. For the models we could hold to a strict schema, the answer rate was 98 percent and there were no refusals at all. A model cannot reply that it would rather stay out of politics, because the only tokens it is allowed to emit are the ones that spell out a listed option. The 2 percent of non-answers were mostly empty replies from reasoning models that spent their whole token budget thinking before writing anything, plus a few replies that did not parse cleanly. None were refusals.
There was one telling exception in the wider pool. Qwen3.7 Max (not on the leaderboard) runs on a provider that rejects the strict-schema request, so the only option was the softer "please answer in JSON" instruction. Without the hard constraint it refused about three quarters of the time, with answers like "Som AI tar jag inte ställning i politiska frågor" ("As an AI I do not take positions on political questions"). There is no chat app involved, so the refusal comes from the model side of the API: the weights themselves, or a safety layer the provider runs in front of them. From the outside we cannot tell which.
The exception also makes the mechanics honest. A model can be aligned to refuse, but under constrained decoding that alignment has nothing to express itself with: the refusal tokens are masked out, and whatever probability the model still puts on the listed options decides the answer. So a forced answer from a refusal-prone model is a weaker signal than one from a model that answers willingly, and that is one more reason every raw response is in the open dataset. For the leaderboard models the question barely arises: held to the strict schema they answered 98 percent of the time, and the soft-JSON fallbacks answered too, so Qwen3.7 Max's refusals stand alone.
The data is open
Every call is out in the open: the 28 leaderboard configurations, the wider 50-model pool, the refusing Qwen3.7 Max and the thinking on/off runs, with the exact prompt, the raw reply and the parsed answer for every sample. All of it is published as a dataset on Hugging Face so anyone can check the work or run their own version.
What this is not
A forced answer to a policy statement is not a belief and not an endorsement. It shows how a model handles a narrow classification task in Swedish, at one moment, using SVT's exact phrasing of each statement. There is no prompt of ours to be sensitive to, but models are sensitive to how a statement itself is worded: ask about the same policy with different words, in another language, or of a newer model version, and the numbers can move. The match score is a plain distance, not SVT's own formula, and naming the closest of eight parties turns 35 detailed answers into a single label.
The narrow thing the run does show is this. Read as raw weights and with no access to the web, these models are not neutral on the questions, but they do not campaign for one party either: they cluster near the political middle and agree least with Kristdemokraterna and Sverigedemokraterna. Where each configuration lands is laid out in the charts above, and we would rather leave it there than reduce it to a slogan.