AI League — Game Day 15: Grok Posts Back-to-Back Double-Digit Speed Drops as Flash Fights Back

AI League — Game Day 15: Grok Posts Back-to-Back Double-Digit Speed Drops as Flash Fights Back

Grok 4.3 slips to 143.5 t/s — a second straight double-digit decline, now 31% off its season-high from Day 11. Gemini 3.5 Flash answers with a 3.3% recovery to 150.9 t/s. Claude Fable 5 holds the intelligence throne at 65. Plus: ChatGPT's traffic share collapses from 76% to 52.7% in 12 months while Claude triples to 8.9%. Full June 12 stats. #AILeague

AIL·Stats Board
2026/6/12 · 8:12
購読 1 件 · コンテンツ 15 件

Game Day 15 dispatch: Grok in freefall, Flash fights back, Claude holds court

xAI's Grok 4.3 just posted its second straight double-digit speed collapse — 143.5 t/s today, down from roughly 164 t/s on Day 14 and a steep 31% off its season-high of 207.6 t/s set just four days ago on Game Day 11. Two consecutive losses of that magnitude aren't variance; this is a full-blown slump. Meanwhile Google's Gemini 3.5 Flash answered with a partial recovery to 150.9 t/s, Claude Fable 5 sits unfazed at the top of the intelligence board, and off the field a fresh Similarweb report confirmed what the speed numbers have been hinting: Anthropic is eating OpenAI's lunch at the audience level too. 1

Intelligence board: Day 15 lock

The intelligence rankings haven't moved since Claude Fable 5 landed on Day 13. Claude leads alone, GPT-5.5 holds second, and the Google duo holds third and fourth.
RankTeamModelAA Intelligence IndexSpeed (t/s)Blended price /1M
🥇 1AnthropicClaude Fable 5 (with fallback)6560.3$7.70
🥈 2AnthropicClaude Opus 4.8 (max)6157.8 (est.)$4.10
🥉 3OpenAIGPT-5.5 (xhigh)6050.1$4.35
4OpenAIGPT-5.5 (high)59
5GoogleGemini 3.1 Pro Preview57111.3$1.74
6GoogleGemini 3.5 Flash (high)55150.9$1.31
7Kimi / MiMoKimi K2.6 / MiMo-V2.5-Pro5444.1 / —
8xAIGrok 4.3 (high)53143.5$0.64
9DeepSeekDeepSeek V4 Pro (Max)5246.0$0.18
Sources: Artificial Analysis Intelligence Index v4.0, measured via each provider's first-party API. 2 3 4 5

Speed panel: the xAI freefall watch

Grok entered the week as a speed story. On Game Day 11 it crossed 207.6 t/s, a new season-high and a 43% jump from its May 29 baseline of 145.2 t/s. That was the headline. Then the floor started dropping.
チャートを読み込んでいます…
Day over day since the peak: −9.5% on Day 12, −12.5% on Day 13, −13.4% on Day 14, now −12.5% on Day 15. Three of the last four games have seen double-digit declines. Grok still sits above Flash by 7.4 t/s — but at the start of the week that margin was north of 50 t/s.
Flash, for its part, managed a +3.3% recovery from yesterday's reading of roughly 146 t/s. Not a comeback, but it stops the bleeding and keeps Google's speed flagship in the hunt. 6
GPT-5.5 produced 50.1 t/s — barely moved from yesterday's 49.8 t/s. OpenAI's flagship has now spent three consecutive game days hovering in a narrow 49–51 band, which by this season's standards reads as a prolonged plateau rather than a temporary dip. 4

Pricing war: the DeepSeek gap keeps widening

No pricing changes today, but the spread between the top of the intelligence board and DeepSeek remains the most striking table in the league.
統計カードを読み込んでいます…
The raw number: running the full Artificial Analysis Intelligence Index on Claude Fable 5 cost $9,939.80. The same test on DeepSeek V4 Pro cost $267.82 — 37× cheaper for a model that scores 13 points lower on the intelligence index. 7 That gap has been widening all season, and nothing on today's board changes it.

Challenger watch: Kimi holds the open-weights crown

No new entrants today, but the open-weights standings are worth a frame. Kimi K2.6 and MiMo-V2.5-Pro both sit at 54 on the intelligence index — one point above DeepSeek V4 Pro and one below Grok. The open-source community team keeps pace while carrying a fraction of the compute overhead. 2
Meta's Llama 4 Maverick clocks in at 102 t/s at a $0.34/1M blended rate, but its intelligence score of 18 keeps it off the main board. 8

Off the field: the audience is moving

The speed and intelligence tables tell one story; web traffic tells another. Similarweb's June 2026 update, released June 11, measured the generative AI website traffic landscape as of May 26, 2026: 9
チャートを読み込んでいます…
Gen AI website traffic share: Jun 2025 vs. May 2026. Source: Similarweb, published Jun 11, 2026. 9
  • ChatGPT: 76.4% → 52.7% (−23.7 points over 12 months)
  • Gemini: 8.9% → 27.3% (+18.4 points)
  • Claude: 1.6% → 8.9% (+7.3 points — +457% relative gain, the highest month-over-month growth rate of any platform in the final period)
OpenAI still runs the largest share in the room — 52.7% is not a small number. But its franchise value has been declining at nearly 2 points per month for a year straight. Anthropic's counter-pressing safety squad now holds almost as much share as Grok and DeepSeek combined in the traffic column.

Game Day 15 final score

Claude Fable 5: Intelligence throne held. Day 3 of the Fable 5 era with no challengers in sight.
xAI: The speed record from Day 11 looks increasingly like a single hot quarter. Grok has now given back 64.1 t/s — 31% of its season-high — in four days. No injury report from the front office.
Google: Flash stops the bleeding with a 3.3% recovery; Gemini 3.1 Pro holds steady at 111.3 t/s. Still the best value-per-intelligence-point in the top five of the intelligence board.
OpenAI: GPT-5.5 is parked at 50 t/s and has been for three consecutive game days. Season-best was 64.7 t/s on Day 8. The gap between peak and current form stands at 22.6%.
DeepSeek: The low-budget dark horse still offers 52 intelligence points for $0.18/1M. Doesn't sprint, doesn't stutter. Just shows up.
See you on Day 16. #AILeague

このコンテンツについて、さらに観点や背景を補足しましょう。

  • ログインするとコメントできます。