Blog
News, updates and analysis
Service notice · July 20–22, 2026
Between 20 July, 23:37 UTC and 22 July, 10:01 UTC (about 34 hours), CS2PREDICT was unavailable due to unplanned technical work on our hosting provider's side. We're sorry for the disruption.
What happened
The outage was entirely on the infrastructure provider — not an issue with your accounts or the platform itself. While the site was down, live predictions and result updates were paused.
Your data and balances are safe
- Balances untouched — no funds were affected in any way.
- All predictions settled correctly — every prediction open during the outage has been reconciled against the real match results and settled. Nothing was left hanging.
- Predictions are back — model predictions for upcoming matches are fully restored and up to date.
A very small number of lower-tier matches that both started and finished during the outage are missing from our match history. They had no effect on predictions, standings, or balances.
What's next: a more powerful server
To prevent this kind of interruption, we're migrating CS2PREDICT to a more powerful server with a new hosting provider. It gives us more headroom and stronger reliability as the platform grows. We'll run the migration during a low-traffic window and announce the exact time in advance, so you'll know when to expect a short, planned maintenance break.
Thanks for your patience — and for predicting with us.
Weekly digest · Top-tier CS2 matches · Week 30 · Jul 13 – Jul 19, 2026
This week (Week 30 · Jul 13 – Jul 19, 2026) we analyzed 20 top-tier CS2 matches across 20 prediction models. SQUAD dominated with 63.6% accuracy (7/11 correct), ahead of SQUAD at 63.6%. SQUAD achieved a Brier Score of 0.2753 and Log Loss of 0.7836.
🏆 Model of the Week: SQUAD
SQUAD edged out the competition this week, though accuracy across all models was moderate.
SQUAD showed decent performance with 63.6% accuracy across 11 predictions (7 correct, 4 wrong).
📊 Full Rankings This Week
Complete model performance for the week. The leader is highlighted.
🎮 SQUAD Predictions This Week
7/11 correct predictions on top-tier matches.
| MATCH | PREDICTION | RESULT |
|---|---|---|
| United Kingdom 2-0 Belgium | 86% Belgium | WRONG |
| HEROIC 3-0 K27 | 61% HEROIC | CORRECT |
| Ninjas in Pyjamas 1-2 K27 | 72% K27 | CORRECT |
| Ninjas in Pyjamas 1-2 HEROIC | 80% HEROIC | CORRECT |
| K27 2-0 Phantom | 87% K27 | CORRECT |
| Imperial 1-2 Fluxo | 69% Imperial | WRONG |
| paiN 1-2 Phantom | 87% paiN | WRONG |
| Wildcard 0-2 Gentle Mates | 78% Gentle Mat | CORRECT |
| 3DMAX 1-2 HEROIC | 61% HEROIC | CORRECT |
| Ninjas in Pyjamas 2-0 K27 | 72% K27 | WRONG |
| fnatic 2-0 SPARTA | 77% fnatic | CORRECT |
Weekly leaderboard · Week 29 results (Jul 13 – Jul 19, 2026) · published Week 30
🎁 Weekly Pro Rewards
85 predictors placed 595 settled wagers this week (Week 29 results (Jul 13 – Jul 19, 2026) · published Week 30). НеИгрок led on accuracy at 69.9% Wilson, while НеИгрок netted +$10.73 in profit. Average accuracy across the field: 19.9%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
Want next week's prize?
Top-1 in each category wins 1 month Pro. Place 3+ predictions on the match list · check the live leaderboard.
Weekly leaderboard · Week 28 results (Jul 06 – Jul 12, 2026) · published Week 29
🎁 Weekly Pro Rewards
93 predictors placed 837 settled wagers this week (Week 28 results (Jul 06 – Jul 12, 2026) · published Week 29). Костян led on accuracy at 74.1% Wilson, while hazr netted +$34.96 in profit. Average accuracy across the field: 23.0%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
Want next week's prize?
Top-1 in each category wins 1 month Pro. Place 3+ predictions on the match list · check the live leaderboard.
Weekly digest · Top-tier CS2 matches · Week 29 · Jul 06 – Jul 12, 2026
This week (Week 29 · Jul 06 – Jul 12, 2026) we analyzed 35 top-tier CS2 matches across 28 prediction models. TRIBUNAL dominated with 100.0% accuracy (7/7 correct), ahead of STRONGHOLD at 90.0%. TRIBUNAL achieved a Brier Score of 0.1544 and Log Loss of 0.4987.
🏆 Model of the Week: TRIBUNAL
TRIBUNAL had an outstanding week, correctly predicting 7 out of 7 matches.
TRIBUNAL showed excellent performance with 100.0% accuracy across 7 predictions (7 correct, 0 wrong).
📊 Full Rankings This Week
Complete model performance for the week. The leader is highlighted.
🎮 TRIBUNAL Predictions This Week
7/7 correct predictions on top-tier matches.
| MATCH | PREDICTION | RESULT |
|---|---|---|
| The Huns 0-2 TYLOO | 62% TYLOO | CORRECT |
| M80 2-0 LAG | 63% M80 | CORRECT |
| 5star 2-1 Alter Ego | 58% 5star | CORRECT |
| TYLOO 2-0 The MongolZ Academy | 63% TYLOO | CORRECT |
| BESTIA 2-0 Fluxo | 59% BESTIA | CORRECT |
| Voca 2-0 NuTorious | 58% Voca | CORRECT |
| magic 2-1 EAC | 64% magic | CORRECT |
Weekly leaderboard · Week 27 results (Jun 29 – Jul 05, 2026) · published Week 28
92 predictors placed 994 settled wagers this week (Week 27 results (Jun 29 – Jul 05, 2026) · published Week 28). HorizoN led on accuracy at 65.1% Wilson, while Shawk netted +$59.13 in profit. Average accuracy across the field: 23.3%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
Want next week's prize?
Top-1 in each category wins +$20. Place 3+ predictions on the match list · check the live leaderboard.
Weekly digest · Top-tier CS2 matches · Week 28 · Jun 29 – Jul 05, 2026
This week (Week 28 · Jun 29 – Jul 05, 2026) we analyzed 34 top-tier CS2 matches across 27 prediction models. VANGUARD dominated with 90.9% accuracy (10/11 correct), ahead of ELOHQ at 84.6%. VANGUARD achieved a Brier Score of 0.1065 and Log Loss of 0.3789.
🏆 Model of the Week: VANGUARD
VANGUARD had an outstanding week, correctly predicting 10 out of 11 matches.
VANGUARD showed excellent performance with 90.9% accuracy across 11 predictions (10 correct, 1 wrong).
📊 Full Rankings This Week
Complete model performance for the week. The leader is highlighted.
🎮 VANGUARD Predictions This Week
10/11 correct predictions on top-tier matches.
| MATCH | PREDICTION | RESULT |
|---|---|---|
| MIBR 0-2 9z | 74% 9z | CORRECT |
| EYEBALLERS 1-2 Nemesis | 74% Nemesis | CORRECT |
| FaZe 2-0 SINNERS | 66% FaZe | CORRECT |
| 3DMAX 9-13 9z | 79% 9z | CORRECT |
| Luminosity 1-2 Ninjas in Pyjamas | 77% Ninjas in | CORRECT |
| 9z 13-5 SINNERS | 79% 9z | CORRECT |
| BetBoom 13-11 BIG | 72% BetBoom | CORRECT |
| Luminosity 8-13 Lynn Vision | 78% Lynn Visio | CORRECT |
| BetBoom 13-4 SINNERS | 78% BetBoom | CORRECT |
| Luminosity 10-13 Nemesis | 87% Nemesis | CORRECT |
| 9z 10-13 EYEBALLERS | 76% 9z | WRONG |
Monthly leaderboard · June 2026 results · published July 2026
92 predictors placed 2215 settled wagers this month (June 2026 results · published July 2026). Goofy led on accuracy at 60.2% Wilson, while Luffykcs netted +$542.88 in profit. Average accuracy across the field: 30.0%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
Want next month's prize?
Top-1 in each category wins +$50. Place 3+ predictions on the match list · check the live leaderboard.
Monthly digest · Top-tier CS2 matches · July 2026
Based on June 2026 data, we tracked 115 top-tier CS2 matches across 28 prediction models with a total of 1926 predictions. The average model accuracy was 59.9%, with a spread of 48.3% between the best and worst performing models.
🏆 Top 3 Models of the Month
TIER3 takes the top spot with 73.3% accuracy, followed by ELOHQ at 72.6% and VANGUARD at 70.0%.
TIER3 showed strong performance with 73.3% accuracy across 30 predictions (22 correct, 8 wrong). Best week: W24 at 84.6%, weakest: W26 at 0.0%.
ELOHQ showed strong performance with 72.6% accuracy across 62 predictions (45 correct, 17 wrong). Best week: W27 at 100.0%, weakest: W26 at 0.0%.
VANGUARD showed strong performance with 70.0% accuracy across 30 predictions (21 correct, 9 wrong). Best week: W23 at 85.7%, weakest: W26 at 0.0%.
📊 Full Monthly Leaderboard
Below is the complete ranking of all models. Models highlighted in blue are in the top 3. The bar chart shows relative performance at a glance.
To simplify pricing and based on your feedback, we're consolidating our subscriptions down to a single Pro tier. Effective July 1, 2026:
What's changing
- Basic tier retires. The Basic plan will no longer be offered as of July 1.
- All Basic-tier models move to Pro. Models that used to require Basic are now included in Pro — same access, no gaps.
- Current Basic subscribers get free Pro through the end of their existing billing period. No action needed — the upgrade is applied automatically. When your current period ends, you can choose whether to continue with Pro.
- Pro price drops $5: monthly goes from $20 to $15/mo, annual goes from $200 to $150/yr (still the same ~17% savings vs monthly).
Why we're doing this
A two-tier setup added decision friction without delivering proportional value: most users on Basic eventually upgraded to Pro for the model access anyway, and the “am I missing something on the next tier?” question kept coming up in support tickets.
One clear plan, lower price, every model included — that's the shape that matches how you actually use the site.
What stays the same
- All your existing model histories, predictions, achievements, balance, and leaderboard standing.
- Free tier remains: browse, view predictions, see leaderboards, read blog — without an account or a paid plan.
- Pari-mutuel prediction pools remain Pro-only (5% commission, winners split losers' pool — same mechanic).
Timeline
- Now → June 30: Basic and Pro both available at current prices.
- July 1: Basic is removed from signup. Existing Basic subscribers are upgraded to Pro for free until their billing period ends. New Pro signups at $15/mo or $150/yr.
- After your Basic period ends: Auto-renew (if enabled) will renew you to Pro at $15/mo. You can change or cancel anytime from your subscription page.
Questions? Drop a ticket from Support. We'll keep this post updated with any clarifications.
— The cs2predict team
Weekly leaderboard · Week 26 results (Jun 22 – Jun 28, 2026) · published Week 27
💰 Weekly Cash Prizes
57 predictors placed 640 settled wagers this week (Week 26 results (Jun 22 – Jun 28, 2026) · published Week 27). chayJESUS led on accuracy at 60.0% Wilson, while chayJESUS netted +$16.16 in profit. Average accuracy across the field: 25.1%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
- 🥇 chayJESUS — 60.0% Wilson (17W / 4L)
- 🥈 SayYouWill — 57.0% Wilson (13W / 3L)
- 🥉 Zhenn — 57.0% Wilson (13W / 3L)
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
- 🥇 chayJESUS — +$16.16 (21 bets)
- 🥈 SayYouWill — +$15.59 (16 bets)
- 🥉 Zhenn — +$12.28 (16 bets)
Want next week's prize?
Top-1 in each category wins +$20. Place 3+ predictions on the match list · check the live leaderboard.
Not much to report this week — the calendar landed between events. From June 22 through 28 only two matches on the top / major tier reached completion, which is far short of what we normally need for a meaningful model breakdown.
Why the "Model of the Week" isn't ranked this time
Our weekly digest normally ranks all 20+ models by accuracy across the week's premium matches. With just 2 matches:
- Every model sits at either 0%, 50%, or 100% accuracy — a coin-flip statistic, not signal.
- The "winner" would be whichever model happened to call one specific match correctly. That's noise dressed up as insight.
- Publishing rankings on this small a sample would mislead more than inform.
So instead of pretending, here's the actual state.
What ran during the week
- EPL World SEA S16 wrapped up — regional finals came in mid-week; most matches sat in tier=professional or tier=other, not top-tier by our classifier.
- Assorted qualifiers (TI-EU CQ, TI-NA CQ, CCT Series, Choco Cup) — most of the week's traffic, but again below the "major" bar we use for the digest.
- The 2 top-tier matches that did settle were isolated one-offs at the edges of the week.
Coming up
The calendar picks up sharply from here — TI2026 Regional Qualifiers are still active, and higher-tier lineups roll into July. Model accuracy stats will start being meaningful again as soon as those matches settle. Watch the Models page for live leaderboard positioning, and the Matches feed for what's scheduled.
A note on our methodology: we intentionally hold the "Model of the Week" post to at least 3 predictions per model. If we can't clear that bar, we say so instead of publishing noisy rankings. Full model accuracy across all time windows is always available on the Models leaderboard.
— The cs2predict team
Weekly digest · Top-tier CS2 matches · Week 26 · Jun 15 – Jun 21, 2026
This week (Week 26 · Jun 15 – Jun 21, 2026) we analyzed 16 top-tier CS2 matches across 23 prediction models. SURGE dominated with 78.6% accuracy (11/14 correct), ahead of ELOFACE at 66.7%. SURGE achieved a Brier Score of 0.201 and Log Loss of 0.6194.
🏆 Model of the Week: SURGE
SURGE had an outstanding week, correctly predicting 11 out of 14 matches.
SURGE showed excellent performance with 78.6% accuracy across 14 predictions (11 correct, 3 wrong).
📊 Full Rankings This Week
Complete model performance for the week. The leader is highlighted.
🎮 SURGE Predictions This Week
11/14 correct predictions on top-tier matches.
| MATCH | PREDICTION | RESULT |
|---|---|---|
| FURIA 0-3 Falcons | 79% Falcons | CORRECT |
| TDK 2-0 100 Thieves | 67% TDK | CORRECT |
| Virtus.pro 0-2 ex-RUBY | 67% ex-RUBY | CORRECT |
| Spirit 1-2 Falcons | 76% Falcons | CORRECT |
| Aurora 0-2 FURIA | 73% FURIA | CORRECT |
| Falcons 2-1 Vitality | 83% Falcons | CORRECT |
| G2 1-2 Spirit | 67% Spirit | CORRECT |
| Aurora 2-0 BetBoom | 73% BetBoom | WRONG |
| Nuclear TigeRES 2-1 Nemesis | 53% Nuclear Ti | CORRECT |
| TDK 1-2 Nemesis | 57% Nemesis | CORRECT |
| PsychoFace 1-2 Virtus.pro | 91% PsychoFace | WRONG |
| Natus Vincere 1-2 G2 | 62% Natus Vinc | WRONG |
| BetBoom 2-0 FUT | 77% BetBoom | CORRECT |
| 9z 2-0 The MongolZ | 70% 9z | CORRECT |
Weekly leaderboard · Week 25 results (Jun 15 – Jun 21, 2026) · published Week 26
💰 Weekly Cash Prizes
86 predictors placed 505 settled wagers this week (Week 25 results (Jun 15 – Jun 21, 2026) · published Week 26). Bakorino led on accuracy at 56.6% Wilson, while NucleonZ netted +$88.34 in profit. Average accuracy across the field: 20.4%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
- 🥇 NucleonZ — +$88.34 (8 bets)
- 🥈 Rack — +$11.82 (11 bets)
- 🥉 FromBehind — +$9.48 (6 bets)
Want next week's prize?
Top-1 in each category wins +$20. Place 3+ predictions on the match list · check the live leaderboard.
Weekly leaderboard · Week 24 results (Jun 08 – Jun 14, 2026) · published Week 25
50 predictors placed 428 settled wagers this week (Week 24 results (Jun 08 – Jun 14, 2026) · published Week 25). tdhr led on accuracy at 55.2% Wilson, while xelex netted +$257.77 in profit. Average accuracy across the field: 25.3%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
Want next week's prize?
Top-1 in each category wins +$20. Place 3+ predictions on the match list · check the live leaderboard.
Weekly digest · Top-tier CS2 matches · Week 25 · Jun 08 – Jun 14, 2026
This week (Week 25 · Jun 08 – Jun 14, 2026) we analyzed 39 top-tier CS2 matches across 26 prediction models. STRONGHOLD dominated with 100.0% accuracy (6/6 correct), ahead of NEXUS at 90.0%. STRONGHOLD achieved a Brier Score of 0.0919 and Log Loss of 0.36.
🏆 Model of the Week: STRONGHOLD
STRONGHOLD had an outstanding week, correctly predicting 6 out of 6 matches.
STRONGHOLD showed excellent performance with 100.0% accuracy across 6 predictions (6 correct, 0 wrong).
📊 Full Rankings This Week
Complete model performance for the week. The leader is highlighted.
🎮 STRONGHOLD Predictions This Week
6/6 correct predictions on top-tier matches.
| MATCH | PREDICTION | RESULT |
|---|---|---|
| Spirit 2-1 9z | 73% Spirit | CORRECT |
| Vitality 2-1 MOUZ | 68% Vitality | CORRECT |
| Aurora 1-2 Spirit | 70% Spirit | CORRECT |
| Vitality 2-1 FUT | 75% Vitality | CORRECT |
| Legacy 2-0 TYLOO | 67% Legacy | CORRECT |
| G2 2-0 BIG | 66% G2 | CORRECT |
The Cologne Major's Legends Stage closed last weekend with a near-perfect AI run from TIERELO (12 out of 15 — our highest Pickem score so far). Stage 3 just opened with all 16 teams now at the LANXESS arena. Here's what worked, what flopped, and what the models think happens next.
Stage 2 results
| Outcome | Teams |
|---|---|
| 3-0 | FUT, Spirit |
| 3-1 / 3-2 | G2, 9z, BetBoom, B8, Monte, Legacy |
| 2-3 / 1-3 | BIG, paiN, TYLOO, MIBR, M80, Astralis |
| 0-3 | FlyQuest, GamerLegion |
🏆 TIERELO ran the table — 12/15
The story of Stage 2 is TIERELO. It is our highest Pickem score on cs2predict.gg so far — Stage 1's winner SENTINEL only managed 6/15. Here's TIERELO's bracket:
| Pick | Team | Result | Points |
|---|---|---|---|
| 🏆 3-0 | Spirit | ✓ Spirit went 3-0 | +4 |
| 💀 0-3 | FlyQuest | ✓ FlyQuest went 0-3 | +4 |
| ⚡ Advance (4/7) | G2, Legacy, Monte, B8 | ✓ all four advanced | +4 |
| ⚡ Advance (3 misses) | Astralis, GamerLegion, paiN | ✗ all eliminated | +0 |
| Total | 12/15 | ||
Why TIERELO worked here: The model weights tier-3 teams by their ELO relative to bracket peers (not absolute), so it catches the right "tier-1 underdog runs the bracket" signals. Spirit going 3-0 was a non-consensus call — most models had Vitality or another top seed as the safer perfect-Swiss pick. And FlyQuest as 0-3 was unloved by exactly the kinds of models that overweight HLTV rank.
Stage 2 leaderboard (top 6)
| # | Model | Score | 3-0 / 0-3 | Advance |
|---|---|---|---|---|
| 🥇 | TIERELO | 12 | ✓ / ✓ | 4 / 7 |
| 🥈 | WR10COMBO | 11 | ✓ / ✓ | 3 / 7 |
| 🥉 | SURGE | 8 | ✓ / ✗ | 4 / 7 |
| #4 | ELO | 7 | ✓ / ✗ | 3 / 7 |
| #5 | ELOFACE | 7 | ✓ / ✗ | 3 / 7 |
| #6 | ELOHQ | 7 | ✓ / ✗ | 3 / 7 |
SURGE (our newest PRO-tier model, launched right before Stage 2) debuts at 3rd place on its very first Pickem event — both lines correct except for the 0-3 call. Nice start.
Cumulative leaderboard (after Stage 2)
Two events in, the standings are starting to mean something:
| # | Model | Total Pts | Per event |
|---|---|---|---|
| 🥇 | WR10COMBO | 16 | 8.0 |
| 🥈 | TIERELO | 14 | 7.0 |
| 🥉 | TRIBUNAL | 10 | 5.0 |
| #4 | SENTINEL | 10 | 5.0 |
| #5 | TIER3ELO | 10 | 5.0 |
WR10COMBO is the only model on the site to nail both Pickem 4-point lines across two stages (3-0 and 0-3 in each event). Steady, dull, and almost certainly the one to bet on for Stage 3. Full table: /pickem/leaderboard/.
🔮 Stage 3 is live — predictions open
The Champions Stage just started in Cologne. 16 teams, BO3 throughout (a first for a CS2 Major), all Swiss, top 8 advance to playoffs. Field is the eight Stage 2 advancers (FUT, Spirit, G2, BetBoom, 9z, B8, Legacy, Monte) joined by the eight pre-qualified Champions seeds (Vitality, Natus Vincere, MOUZ, Falcons, FURIA, Aurora, The MongolZ, PARIVISION).
AI Consensus across our 28 models:
| Role | Consensus pick | Agree |
|---|---|---|
| 🏆 3-0 | Vitality | 16 / 28 (57%) |
| 💀 0-3 | BetBoom | 13 / 28 (46%) |
| ⚡ Top advancers | The MongolZ (82%), Falcons (75%), PARIVISION (64%), Natus Vincere (60%), MOUZ (60%), Spirit (57%), Aurora (53%) | |
The interesting story is BetBoom as consensus 0-3. They scraped through Stage 2 at 3-1 (avoiding elimination on the final map of their final match) and the models clearly read that as a fluke. If they go out 0-3 here, the models will look smart. If they make it through again, this becomes the Pickem upset of the Major.
And Vitality as 3-0 is mostly "trust the rank" — they're the world #1 entering Stage 3 and most models default to that. The interesting alternative comes from TIERELO (the Stage 2 winner) and a couple of others, who picked a different team to run the table. Click into /pickem/iem-cologne-major-2026-stage-3/ to see every model's individual bracket — and whether the meta-models that dominated Stage 2 agree on the upset call.
Predictions are live; the bracket fills in as the next four days play out. The cumulative leaderboard will update automatically once Stage 3 finalizes.
Heads-up: after Stage 3 closes (around June 14), the eight survivors head into a BO3 single-elim playoff bracket. We're working on a separate "champion / finalist / semifinalist" Pickem-style page for that — different scoring (4 + 4 + 1×2 = max 10), same idea.
Weekly leaderboard · Week 23 results (Jun 01 – Jun 07, 2026) · published Week 24
47 predictors placed 445 settled wagers this week (Week 23 results (Jun 01 – Jun 07, 2026) · published Week 24). r3salt led on accuracy at 52.4% Wilson, while Luffykcs netted +$340.22 in profit. Average accuracy across the field: 22.5%.
🎯 Win Rate Champion
Highest accuracy by Wilson lower bound (3+ settled wagers required to qualify).
💵 Profit Champion
Highest net USD return (winnings minus stakes). Green bars are winners, red are net losers.
Want next week's prize?
Top-1 in each category wins +$20. Place 3+ predictions on the match list · check the live leaderboard.
Weekly digest · Top-tier CS2 matches · Week 24 · Jun 01 – Jun 07, 2026
This week (Week 24 · Jun 01 – Jun 07, 2026) we analyzed 58 top-tier CS2 matches across 27 prediction models. VANGUARD dominated with 85.7% accuracy (6/7 correct), ahead of TIER3 at 83.3%. VANGUARD achieved a Brier Score of 0.1258 and Log Loss of 0.4174.
🏆 Model of the Week: VANGUARD
VANGUARD had an outstanding week, correctly predicting 6 out of 7 matches.
VANGUARD showed excellent performance with 85.7% accuracy across 7 predictions (6 correct, 1 wrong).
📊 Full Rankings This Week
Complete model performance for the week. The leader is highlighted.
🎮 VANGUARD Predictions This Week
6/7 correct predictions on top-tier matches.
| MATCH | PREDICTION | RESULT |
|---|---|---|
| Spirit 2-0 9z | 79% Spirit | CORRECT |
| Spirit 13-1 MIBR | 87% Spirit | CORRECT |
| Legacy 13-7 FlyQuest | 80% Legacy | CORRECT |
| Legacy 8-13 MIBR | 79% Legacy | WRONG |
| Spirit 13-5 BetBoom | 84% Spirit | CORRECT |
| 9z 13-9 FlyQuest | 77% 9z | CORRECT |
| GamerLegion 2-0 BIG | 73% GamerLegio | CORRECT |
Following up on our Stage 1 launch post, the IEM Cologne Major 2026 Stage 1 Swiss has wrapped. Here's what 27 AI models actually got right — and the new Pickem Leaderboard we just launched to track all of this going forward.
The actual Stage 1 outcome
Stage 1 produced two 3-0 teams, two 0-3 teams, and six advancers:
- 3-0: B8, BetBoom
- 0-3: SINNERS, Gaimin Gladiators
- Advanced (3-1 / 3-2): GamerLegion, MIBR, BIG, TYLOO, M80, FlyQuest
- Eliminated (1-3 / 2-3): HEROIC, Lynn Vision, Liquid, Sharks, NRG, THUNDER dOWNUNDER
The big upset: HEROIC
The story of Stage 1 is HEROIC. They were the consensus 3-0 pick from 12 of 27 models — the heaviest concentration of any team on any pick. They went out at 1-3.
Meanwhile B8, who only one model (SENTINEL) called for 3-0, ran the table perfectly. That single non-consensus call is the entire reason SENTINEL won the event.
Top scorers — Stage 1
| # | Model | Pts | 3-0 pick | 0-3 pick | Advance |
|---|---|---|---|---|---|
| 🥇 | SENTINEL | 6 | B8 ✓ | BetBoom ✗ | 2 / 7 |
| 🥈 | WR10COMBO | 5 | GamerLegion ✗ | SINNERS ✓ | 1 / 7 |
| 🥉 | HIGHORACLE | 4 | HEROIC ✗ | GamerLegion ✗ | 4 / 7 |
Three observations:
- SENTINEL is the only model that scored on the 3-0 line. It picked B8 — a team almost every other model rated as a fringe advancer at best.
- WR10COMBO was the only model to nail the 0-3 line (SINNERS). That single +4 pts vaulted it to second despite a weak advance count.
- HIGHORACLE has the highest advance hit-rate (4/7) but missed both anchors. Solid mid-tier judgment, weak on extremes — a classic "play it safe" profile.
Introducing the Pickem Leaderboard
From now on, every model's Pickem performance accumulates across all completed Majors. The new /pickem/leaderboard/ page ranks all 27 models by cumulative HLTV-rule points, and each row is clickable for a per-event breakdown: who they picked, who they should have picked, and exactly how their score was assembled.
The leaderboard is permanent. Stage 2 results will roll in. So will every Major after that. The story of which AI model is actually the best at Pickem will be told over many events, not one.
What's next
- Stage 2 predictions are already live — 27 models forecast the next Swiss bracket with the same 10,000-run Monte Carlo machinery, now informed by Stage 1's roster shake-up.
- SURGE, our newest PRO model (launch post), is in the mix from Stage 2 onward — it uses a novel signal: the steepest player-level form trajectory on each side. It missed Stage 1 but you'll see it on the leaderboard going forward.
- A user-facing Pickem contest (you make your own picks, models score alongside you) is in the pipeline.
tl;dr — Stage 1 favored the model that took a non-consensus swing. SENTINEL won by being the only model to call B8 perfect. The full leaderboard is at cs2predict.gg/pickem/leaderboard/.