UTC05:21ZNowSIREN 456.6/100Delta+0.5upsince 03:34ZObservedSources15/15reportingStatusOPERATIONAL

Who is winning the AI race?

★ FEATURED · THE NATIONS RACE 🇺🇸 USA 98%3–22.1% CHINA 🇨🇳 Market odds that the best AI model is American vs Chinese, plus labs and electricity. USA vs China: who is winning the race to superintelligence? →

Who is winning, and how sure is the money? 8 frontier labs and the people running them, ranked by a live prediction market and annotated with what each one actually shipped. Every number links to the thing it came from.

These are market odds and observable activity. They are not a SIREN prediction. Players are ranked by the live Polymarket probability that they hold the best AI model at the furthest-out resolution date still trading. This is the market’s ranking, not SIREN’s. Nothing on this page is a SIREN forecast, and none of it feeds the SIREN index. They are also not the Artificial Analysis Intelligence Index — that ranks models on independent quality, speed and cost benchmarks. This page ranks labs on live prediction-market prices.

Compiled
Ranking market
Which company has best AI model end of 2026?
Resolves
Behind it
$2.52M across 15 legs
Cross-check
Kalshi 12 contracts, 421k

How are the odds split across the board?The whole board

15 legs are trading on Which company has best AI model end of 2026?, and they are mutually exclusive: exactly one resolves YES. Their prices sum to 1.0075 rather than to 1.0000 — that spread is the bid/ask on $2.52M of lifetime volume. Segment widths above are normalised by that sum; the percentages printed are the raw prices.

Trade this market on Polymarket → Prediction markets are restricted in some places. Not financial advice.

How do the labs sit on one axis?The field

Every lab on the board, on one absolute 0–100% axis. The dot is the price now; the hollow ring is where that leg traded seven days ago. Nothing is normalised to the leader — one lab really is most of this market.

Frontier labs ranked by the live year-end market probabilityEach dot is the live price of that lab's leg, on an absolute 0 to 100 per cent axis; the hollow ring beside it is where the leg traded seven days ago. Anthropic 44.5%, up +5.0 points. Google DeepMind 39.5%, down −2.0 points. OpenAI 7.8%, up +1.1 points. Meta AI 4.5%, up +1.5 points. xAI 1.8%, down −0.4 points. Alibaba Qwen 0.55%, no seven-day reference. DeepSeek 0.45%, down −0.1 points. Mistral 0.25%, no seven-day reference.SHARE OF THE YEAR-END MARKET● NOW ○ 7D AGO0%0%50%50%100%100%1Anthropic44.5%▲ +5.0 pts2Google DeepMind39.5%▼ −2.0 pts3OpenAI7.8%▲ +1.1 pts4Meta AI4.5%▲ +1.5 pts5xAI1.8%▼ −0.4 pts6Alibaba Qwen0.55%no 7d ref7DeepSeek0.45%▼ −0.1 pts8Mistral0.25%no 7d refFrontier labs ranked by the live year-end market probabilityEach dot is the live price of that lab's leg, on an absolute 0 to 100 per cent axis; the hollow ring beside it is where the leg traded seven days ago. Anthropic 44.5%, up +5.0 points. Google DeepMind 39.5%, down −2.0 points. OpenAI 7.8%, up +1.1 points. Meta AI 4.5%, up +1.5 points. xAI 1.8%, down −0.4 points. Alibaba Qwen 0.55%, no seven-day reference. DeepSeek 0.45%, down −0.1 points. Mistral 0.25%, no seven-day reference.SHARE OF THE YEAR-END MARKET● NOW ○ 7 DAYS AGO0%0%25%25%50%50%75%75%100%100%1Anthropic44.5%▲ +5.0 pts2Google DeepMind39.5%▼ −2.0 pts3OpenAI7.8%▲ +1.1 pts4Meta AI4.5%▲ +1.5 pts5xAI1.8%▼ −0.4 pts6Alibaba Qwen0.55%no 7d ref7DeepSeek0.45%▼ −0.1 pts8Mistral0.25%no 7d refFrontier labs ranked by the live year-end market probabilityEach dot is the live price of that lab's leg, on an absolute 0 to 100 per cent axis; the hollow ring beside it is where the leg traded seven days ago. Anthropic 44.5%, up +5.0 points. Google DeepMind 39.5%, down −2.0 points. OpenAI 7.8%, up +1.1 points. Meta AI 4.5%, up +1.5 points. xAI 1.8%, down −0.4 points. Alibaba Qwen 0.55%, no seven-day reference. DeepSeek 0.45%, down −0.1 points. Mistral 0.25%, no seven-day reference.SHARE OF THE YEAR-END MARKET● NOW ○ 7 DAYS AGO0%0%10%10%20%20%30%30%40%40%50%50%60%60%70%70%80%80%90%90%100%100%1Anthropic44.5%▲ +5.0 pts2Google DeepMind39.5%▼ −2.0 pts3OpenAI7.8%▲ +1.1 pts4Meta AI4.5%▲ +1.5 pts5xAI1.8%▼ −0.4 pts6Alibaba Qwen0.55%no 7d ref7DeepSeek0.45%▼ −0.1 pts8Mistral0.25%no 7d ref
Anthropic holds 44.5% of the year-end market — 5.0 points clear of Google DeepMind at 39.5%. The hollow ring on each row is where that leg traded seven days ago; a row with no ring has no week-old price to difference against.

How did each lab change over seven days?Seven days

Two observations per lab and a line joining them. There is no price history in race.json, so there is no curve here to draw and none is invented.

Where each lab’s year-end odds moved over seven daysTwo observations per lab, joined by a straight line: the price the venue published seven days ago and the price it publishes now. Nothing between those two points was recorded and nothing between them is drawn as data. Anthropic went from 39.5% to 44.5%, up +5.0 points. Google DeepMind went from 41.5% to 39.5%, down −2.0 points. OpenAI went from 6.7% to 7.8%, up +1.1 points. Meta AI went from 3.0% to 4.5%, up +1.5 points. xAI went from 2.2% to 1.8%, down −0.4 points. DeepSeek went from 0.50% to 0.45%, down −0.1 points. No seven-day reference, and so not drawn: Alibaba Qwen, Mistral.THE WHOLE FIELD · 0–50%0%25%50%7 DAYS AGONOWAnthropic 44.5%Google DeepMind 39.5%OpenAI 7.8%Meta AI 4.5%xAI 1.8%DeepSeek 0.45%NO 7D REFERENCE, NOT DRAWN: Alibaba Qwen, MistralWhere each lab’s year-end odds moved over seven daysTwo observations per lab, joined by a straight line: the price the venue published seven days ago and the price it publishes now. Nothing between those two points was recorded and nothing between them is drawn as data. Anthropic went from 39.5% to 44.5%, up +5.0 points. Google DeepMind went from 41.5% to 39.5%, down −2.0 points. OpenAI went from 6.7% to 7.8%, up +1.1 points. Meta AI went from 3.0% to 4.5%, up +1.5 points. xAI went from 2.2% to 1.8%, down −0.4 points. DeepSeek went from 0.50% to 0.45%, down −0.1 points. No seven-day reference, and so not drawn: Alibaba Qwen, Mistral.THE WHOLE FIELD · 0–50%0%10%20%30%40%50%7 DAYS AGONOWAnthropic 44.5%Google DeepMind 39.5%OpenAI 7.8%Meta AI 4.5%xAI 1.8%DeepSeek 0.45%NO 7D REFERENCE, NOT DRAWN: Alibaba Qwen, MistralWhere each lab’s year-end odds moved over seven daysTwo observations per lab, joined by a straight line: the price the venue published seven days ago and the price it publishes now. Nothing between those two points was recorded and nothing between them is drawn as data. Anthropic went from 39.5% to 44.5%, up +5.0 points. Google DeepMind went from 41.5% to 39.5%, down −2.0 points. OpenAI went from 6.7% to 7.8%, up +1.1 points. Meta AI went from 3.0% to 4.5%, up +1.5 points. xAI went from 2.2% to 1.8%, down −0.4 points. DeepSeek went from 0.50% to 0.45%, down −0.1 points. No seven-day reference, and so not drawn: Alibaba Qwen, Mistral.THE WHOLE FIELD · 0–50%0%10%20%30%40%50%7 DAYS AGONOWAnthropic 44.5%Google DeepMind 39.5%OpenAI 7.8%Meta AI 4.5%xAI 1.8%DeepSeek 0.45%NO 7D REFERENCE, NOT DRAWN: Alibaba Qwen, Mistral
Biggest seven-day move: Anthropic, up +5.0 points, from 39.5% to 44.5%. Two observations per lab — now, and the price implied by the venue’s own 7-day change. data/race.json holds no series, so the line between them is a join and not a path: nothing was recorded in between and nothing in between is drawn.

Who leads the AI race board?

The fill behind each row is the live year-end probability on an absolute 0–100% scale, never normalised to the leader, and the figure at the right edge is the same number. The tick at the track's edge is where that leg traded seven days ago. Under every row: the second venue, what the lab shipped in 30 days, and its share of the news window.

The AI race: 8 frontier labs ranked by live market probability, with shipping activity and news mindshare. Compiled 2026-10-15 05:24:48 UTC.

  1. Rank 1 What has SIREN scored about Anthropic?Dario Amodei +5.00 points, up, over seven days44.5% of the year-end market, 39.50% a week ago
    Market
    $345k traded39.50% a week ago
    Cross-check
    today30.00%kalshino market
    Shipping, 30d
    gh≥40hfnot usedor6
    Mindshare
    10.3%82 of 800 items+2.3 points of corpus share, up

    why Market leader at 44.5%, +5.0 points clear of Google DeepMind. +5.0 points over 7 days.

    every signal we have on Anthropic
    • +5.0 points over 7 days.
    • Kalshi lists no contract for them.
    • Named in 10.3% of the 800-item news window (+2.3 points across the window).
    • ≥40 GitHub releases, 6 OpenRouter listings in 30 days.

    Loudness: darioamodei.com publishes essays but serves no feed: /rss and /feed.xml both returned HTTP 404 (probed 2026-09-23).

  2. Rank 2 What has SIREN scored about Google DeepMind?Demis Hassabis −2.00 points, down, over seven days39.5% of the year-end market, 41.50% a week ago
    Market
    $290k traded41.50% a week ago
    Cross-check
    today68.30%kalshino market
    Shipping, 30d
    gh≥31hf0or3
    Mindshare
    3.3%26 of 800 items−1.1 points of corpus share, down

    why 39.5%, −5.0 points against Anthropic. −2.0 points over 7 days.

    every signal we have on Google DeepMind
    • −2.0 points over 7 days.
    • Kalshi lists no contract for them.
    • Named in 3.3% of the 800-item news window (−1.1 points across the window).
    • ≥31 GitHub releases, 3 OpenRouter listings in 30 days.

    Loudness: No personal feed. hassabis.com serves a 114-byte JavaScript redirect, not a document (probed 2026-09-23).

  3. Rank 3 What has SIREN scored about OpenAI?Sam Altman +1.15 points, up, over seven days7.80% of the year-end market, 6.65% a week ago
    Market
    $213k traded6.65% a week ago
    Cross-check
    today0.60%kalshi18%
    Shipping, 30d
    gh≥34hf0or12
    Mindshare
    12.3%98 of 800 items+3.0 points of corpus share, up

    why 7.8%, −36.7 points against Anthropic. +1.1 points over 7 days.

    every signal we have on OpenAI
    • +1.1 points over 7 days.
    • Kalshi's separate "top-ranked this year" contract has them at 18%.
    • Named in 12.3% of the 800-item news window (+3.0 points across the window).
    • ≥34 GitHub releases, 12 OpenRouter listings in 30 days.
    • Sam Altman's own blog has been silent since 2026-04-10.
  4. Rank 4 What has SIREN scored about Meta AI?Mark Zuckerberg +1.45 points, up, over seven days4.45% of the year-end market, 3.00% a week ago
    Market
    $182k traded3.00% a week ago
    Cross-check
    today0.45%kalshi9%
    Shipping, 30d
    gh1hf1or1
    Mindshare
    0.9%floor7 of 800 items−1.5 points of corpus share, down

    why 4.5%, −40.1 points against Anthropic. +1.5 points over 7 days.

    every signal we have on Meta AI
    • +1.5 points over 7 days.
    • Kalshi's separate "top-ranked this year" contract has them at 9%.
    • Named in 0.9% of the 800-item news window (−1.5 points across the window).
    • 1 GitHub release, 1 Hugging Face model repo, 1 OpenRouter listing in 30 days.

    Floor, not a count. The published vocabulary matches "Meta AI" but not the bare token "Meta", because "meta-learning" and "meta-analysis" would match it in a corpus that is mostly ML preprints. Headlines that say only "Meta" are not counted here.

    Loudness: Posts on Facebook and Instagram, neither of which exposes a free per-person feed at any documented path.

  5. Rank 5 What has SIREN scored about xAI?Elon Musk −0.35 points, down, over seven days1.85% of the year-end market, 2.20% a week ago
    Market
    $167k traded2.20% a week ago
    Cross-check
    today0.15%kalshi8%
    Shipping, 30d
    gh1hf0or1
    Mindshare
    0.6%5 of 800 items−0.9 points of corpus share, down

    why 1.8%, −42.6 points against Anthropic. 1 GitHub release, 1 OpenRouter listing in 30 days.

    every signal we have on xAI
    • −0.4 points over 7 days.
    • Kalshi's separate "top-ranked this year" contract has them at 8%.
    • Named in 0.6% of the 800-item news window (−0.9 points across the window).
    • 1 GitHub release, 1 OpenRouter listing in 30 days.

    Loudness: Posts on X. Scraping X carries an explicit permanent-suspension penalty under its developer terms, so this cell is empty by policy, not by accident.

  6. Rank 6 What has SIREN scored about Alibaba Qwen?no single principal no ref0.55% of the year-end market
    Market
    $145k tradedweek-old price: no ref
    Cross-check
    today0.05%kalshi1%
    Shipping, 30d
    gh≥10hf3or2
    Mindshare
    0.4%3 of 800 items+0.5 points of corpus share, up

    why 0.5%, −44.0 points against Anthropic. ≥10 GitHub releases, 3 Hugging Face model repos, 2 OpenRouter listings in 30 days.

    every signal we have on Alibaba Qwen
    • Polymarket has no week-old reference for this leg.
    • Kalshi's separate "top-ranked this year" contract has them at 1%.
    • Named in 0.4% of the 800-item news window (+0.5 points across the window).
    • ≥10 GitHub releases, 3 Hugging Face model repos, 2 OpenRouter listings in 30 days.

    Loudness: No principal to follow — see the roster note.

  7. Rank 7 What has SIREN scored about DeepSeek?Liang Wenfeng −0.05 points, down, over seven days0.45% of the year-end market, 0.50% a week ago
    Market
    $142k traded0.50% a week ago
    Cross-check
    today0.05%kalshi5%
    Shipping, 30d
    gh0hf0or0
    Mindshare
    0.3%2 of 800 items+0.3 points of corpus share, up

    why 0.4%, −44.0 points against Anthropic. Nothing published to the three channels we watch in 30 days.

    every signal we have on DeepSeek
    • −0.1 points over 7 days.
    • Kalshi's separate "top-ranked this year" contract has them at 5%.
    • Named in 0.3% of the 800-item news window (+0.3 points across the window).
    • Nothing published to the three channels we watch in 30 days.

    Loudness: Liang Wenfeng maintains no public feed in any language we can fetch.

  8. Rank 8 What has SIREN scored about Mistral?Arthur Mensch no ref0.25% of the year-end market
    Market
    $191k tradedweek-old price: no ref
    Cross-check
    today0.05%kalshi3%
    Shipping, 30d
    gh6hf2or1
    Mindshare
    0.0%0 of 800 items±0.0 points of corpus share, unchanged

    why 0.3%, −44.3 points against Anthropic. 6 GitHub releases, 2 Hugging Face model repos, 1 OpenRouter listing in 30 days.

    every signal we have on Mistral
    • Polymarket has no week-old reference for this leg.
    • Kalshi's separate "top-ranked this year" contract has them at 3%.
    • Named in 0.0% of the 800-item news window (±0.0 points across the window).
    • 6 GitHub releases, 2 Hugging Face model repos, 1 OpenRouter listing in 30 days.

    Loudness: No personal feed. Mistral publishes as a company; Arthur Mensch does not.

How is the race board computed?

The rank
Every player is placed by one number and one number only: the live YES price on their leg of Which company has best AI model end of 2026?, resolving 2026-12-31. That event is a negRisk market — 15 mutually-exclusive legs, exactly one of which resolves YES — so a leg price is a live, money-backed probability for that lab and nothing else. We do not blend the probability with shipping and mindshare into a composite score. A blend would need weights no one can check against an outcome, and the resulting rank would be ours rather than the market's. The other readings on each row are context. They do not move the order.
Cross-check columns
today is the same question at the nearest resolution date (Which company has the best AI model end of October?, 2026-10-31) — who holds the crown right now rather than at year end. The two disagree sharply and they are supposed to. kalshi is a different question on a CFTC-regulated venue: “Which companies will have a top-ranked AI model this year?”, whose legs are independent binaries and do not sum to 1. It lists 12 contracts and carries no leg at all for some of this roster, which is printed as “no market”, never as 0%.
Shipping
Three channels, published side by side and never added together, because they are biased in opposite directions and a sum would hide both. gh counts releases across a fixed basket of that lab's own repositories — which measures release engineering as much as shipping, since some labs auto-cut an SDK tag on every API change and others ship weights and never cut a tag. A ≥ means the feed's ten-entry ceiling was hit and the real number is higher. hf counts model repositories the lab created on the Hugging Face Hub — a vendor-side publication event, and structurally zero for a lab that does not publish weights. or counts new entries in the OpenRouter catalogue. That timestamp is when OpenRouter listed the model, not when the vendor announced it, and it is never labelled as a release date anywhere on this page. All three windows are 30 days.
Mindshare
Share of SIREN's own news corpus naming that lab, computed from the entity list Which AI stories has SIREN scored in this window? already published — the same extractor, so the two surfaces can never disagree about a headline. The window is 800 items spanning 66.0 hours, not the nominal 7 days: the corpus is capped at 800 items and on a busy week the cap binds long before the window does. The trend compares the recent half (657 items) against the prior half (143 items); both must carry at least 20 items or it is reported as too thin rather than computed anyway. Shares do not sum to 100% — one item naming two labs counts for both — and only 324 of 800 items name any lab at all, because most of the corpus is research preprints. This is share of our corpus. It is not share of the internet.
Loudness
The brief asked for the public posting cadence of each principal, from feeds we can legitimately fetch. Applied honestly that leaves almost nothing, and the empty column is the finding. Six of these eight people post on X, and scraping X carries an explicit permanent-suspension penalty under its developer terms — so that door is closed, not hard. Facebook and Instagram expose no free per-person feed. One legitimate feed exists and it is published by the principal himself; where it is absent, the reason is printed on the row rather than replaced with a zero.
Manifold
The only comparable Manifold market carries 1,137 MANA of play money against the floor of 20,000. Published as a measured exclusion, not omitted. The probe is re-run every build rather than asserted once in a comment, so the exclusion cannot quietly go stale.
What this is not
Not a quality ranking. Artificial Analysis publishes independent intelligence, speed and cost leaderboards. SIREN does not copy those scores onto this page. Money-backed odds and shipping counts answer a different question.

What does a blank reading mean?

Formula version 1.0.0. The full method, the market selection rules and the traps are in docs/RACE.md, and every input on this page is in api/race.json.

How healthy are the race sources?

All 8 instruments answered on this run.

What questions do people actually ask?

Who is winning the AI race?

Anthropic leads at 44.5% among 8 frontier labs ranked by a live prediction market. These are market odds and observable activity — not a SIREN prediction and not a race doom score.

How do the labs sit on one axis?

Every lab on the board, on one absolute 0–100% axis. The dot is the price now; the hollow ring is where that leg traded seven days ago. Nothing is normalised to the leader — one lab really is most of this market.

How did each lab change over seven days?

Two observations per lab and a line joining them. There is no price history in race.json, so there is no curve here to draw and none is invented.

How are the odds split across the board?

15 legs are trading on Which company has best AI model end of 2026?, and they are mutually exclusive: exactly one resolves YES. Segment widths are normalised by the sum of live legs; the percentages printed are the raw prices.

Who leads the AI race board?

Anthropic leads at 44.5%. The fill behind each row is the live year-end probability on an absolute 0–100% scale, never normalised to the leader.

How healthy are the race sources?

All 8 instruments answered on this run.

How is the race board computed?

Labs are ranked by a live prediction market and annotated with shipping activity and news mindshare. Every number links to the thing it came from. Not a quality ranking and not a SIREN prediction.

What does a blank reading mean?

A blank mark means we looked and the channel did not answer, or the absence legend applies. A printed zero is the only absence that is a measurement: we looked, the channel answered, and the answer is genuinely zero.

Does SIREN publish an AI race doom score?

No. These are market odds and observable activity. They are not a SIREN prediction. Activity tempo, not probability of harm.

SIREN is free and carries no display ads. Keep it running: Support SIREN on Ko-fi →

News & alerts live · news – · SIREN score (hourly) Feedback ☕ Support𝕏 Post