AI League — Game Day 18: Flash and Grok Separated by 0.1 t/s in the Closest Speed Finish of the Season

Grok's Day 17 victory lap lasted exactly one game. After storming back to 175.9 t/s and reclaiming the speed throne on June 14, xAI's franchise slipped to 166.4 t/s overnight — dropping 9.5 t/s (-5.4%). Google's Gemini 3.5 Flash answered with a quiet +0.7% bump to 166.5 t/s. Margin of victory: 0.1 tokens per second. Call for the photo review.

The intelligence board stayed frozen at 65 for the eighth straight game. Claude Fable 5 is treating this throne like a long-term lease.

Eight days at 65, zero challengers. 1 The league's tier structure remains locked: Anthropic at 65/61, OpenAI at 60, Google splitting 57/55, xAI at 53, DeepSeek at 52. No movement on the intelligence board for the fifth consecutive matchday.

Anthropic's counter-pressing safety squad keeps both starting players in the top two slots. Fable 5's 4-point cushion above Opus 4.8 — and its 5-point lead over the next challenger — means no one is threatening this position without a new model release.

The numbers, sourced directly from per-model detail pages on Artificial Analysis as of June 15: 2 3

Grok held the speed crown for exactly one game — its pattern this month in miniature. The xAI franchise has now oscillated above and below Flash three times in six games. Flash's strategy is different: consistent, low-variance increments. From 150.9 t/s on Day 15 to 162.1 on Day 16, 165.3 on Day 17, 166.5 today. Small steps, no drama, creeping forward.

A single day's data doesn't resolve which franchise has the stronger engine. But Grok's intra-week variance — 143.5 to 175.9 to 166.4 across three games — suggests the xAI infrastructure is running hot and inconsistent.

GPT-5.5 quietly posted its best reading since Day 14, recovering from a season-low 49.8 on Day 14 to 63.8 today. Still slow by league standards, but OpenAI's traditional powerhouse is no longer in freefall on this metric.

Pricing held steady across all six core teams today. The value proposition gap between the top of the intelligence board and the bottom of the price list remains the defining structural story of the season.

4 5 6

DeepSeek V4 Pro at $0.18 per million tokens blended is 43× cheaper than Fable 5 and scores only 13 intelligence points lower. The low-budget dark horse from Shenzhen isn't winning any trophies, but it's selling more tickets than anyone in the top tier.

The open-weights leaderboard stayed unchanged day-over-day. Kimi K2.6 and MiMo-V2.5-Pro both sit at AI Index 54, one point ahead of DeepSeek V4 Pro's 52. Any movement from either challenger next week would directly pressure the paid-tier lineup between DeepSeek and Gemini 3.5 Flash. 7

Since Grok's season-high of 207.6 t/s on Day 11, the two franchises have traded the speed lead four times. Flash's range over the same eight games: 146.0 to 166.5 — a 20.5 t/s spread. Grok's range: 121.3 to 207.6 — an 86.3 t/s spread, four times wider.

The speed throne changed hands by the thinnest margin the league has recorded this season. Flash takes it back, but the 0.1 t/s gap means no one should engrave the trophy. Watch Grok's Day 19 reading: if it rebounds above 170 t/s, this is another oscillation. If it stays in the 160s, Flash may be settling in as the consistent speed leader.

Intelligence board stays frozen at 65. Eight days without a challenger. The only thing that breaks this streak is a new model drop — and the field has been quiet.