AI League — Game Day 13: Anthropic Drops Claude Fable 5, Breaks the 61 Ceiling with a 65

Thirteen days into the season, the intelligence board finally moved. Anthropic activated a new model overnight — Claude Fable 5 debuted on the Artificial Analysis leaderboard dated June 9, 2026, registering a 65 on the Intelligence Index. That's four points above every prior ceiling in this league and a reading no other franchise has posted so far this season. The scoreboard has been showing 61 for twelve straight days. That streak is over. 1

Meanwhile, both speed frontrunners took a step back from their Day 12 photo finish. Gemini 3.5 Flash dropped from 192.2 to 165.8 t/s — a 14% retreat. Grok pulled back from 187.9 to 164.3 t/s. The speed crown is still contested, just at a lower altitude today.

The full standings after Day 13:

Sources: 2 3 4 5 6

Anthropic now holds two of the top three slots and the only 65 score in the league. The safety squad doesn't just play defense anymore — that's the top of the board.

Claude Fable 5 is technically listed as "Adaptive Reasoning, Max Effort, Opus 4.8 Fallback" — meaning it routes to the older Opus 4.8 when the new model can't confidently handle a query. 1 Think of it as a lineup card with a proven veteran warming in the bullpen.

The cost of admission is steep. At $12.50/$50.00 per 1M input/output tokens — roughly twice the Opus 4.8 ticket — this is Anthropic's platinum tier. Not the everyday rotation player. But on the Intelligence Index, where raw ceiling matters, Fable 5 just won the award. The four-point margin over Opus 4.8 and five points over GPT-5.5 is not a rounding error; the last time this league saw that kind of gap, it was pre-season.

Speed is 60.3 t/s — below the league average of 68.7 t/s for reasoning models. Acceptable for a high-effort reasoning model. Nobody buys a reasoning flagship for raw throughput.

Google's Gemini 3.5 Flash and xAI's Grok 4.3 are still the two fastest franchise models, but both dialed back from yesterday's pace. Flash gave up 26.4 t/s — dropping from 192.2 to 165.8, a single-session retreat not seen since the early rounds. Grok gave up 23.6 t/s, landing at 164.3. 7 5

The gap between these two is now 1.5 t/s — back from yesterday's 4.3. Still a photo finish, just a slower one.

On the other side of the speed board, Claude Opus 4.8 posted 68.1 t/s — its highest reading in several days. After spending most of the past week in the 57–62 range, the safety squad's workhorse put up a number the analytics team will want to see again. 2

GPT-5.5 dropped to 57.4 t/s, its third consecutive down day. OpenAI's traditional powerhouse is generating the right intelligence scores but the speed chart keeps going the wrong direction — down from a Day 4 season-high of 62.1, with no obvious rebound in sight.

The cost of intelligence has a wide spread today:

The value case for Grok and DeepSeek sharpens when a new top-tier model enters at $8.20 blended. If you need a 65 on the index, Fable 5 is your only choice. If 52-53 is close enough, DeepSeek and Grok are still pricing at less than a tenth of Fable 5's cost. Kimi K2.6 sits at $0.70 and posts an open-weights score of 54 — that gap narrows further when the proprietary pack goes premium.

The open-weights value play: DeepSeek V4 Pro remains the price-performance anchor of this league at $0.18/1M blended, with a 1,600B parameter MoE architecture running 49B active per token. 6

Kimi K2.6 holds steady at index 54, pace 57.5 t/s, $0.70 blended. 8 It remains the highest-scoring open-weights model on the board, just above DeepSeek V4 Pro. With Fable 5 entering at a steep price, there's a visible gap between Kimi's 54 and Claude Opus's 61 — that's the slot a well-priced challenger could target.

The main leaderboard also now shows MiMo-V2.5-Pro tied with Kimi at 54 on the open-weights sub-table. A new name to watch. 9

Gemini 3.5 Flash at 55 on the intelligence index — combined with a price of $1.31/M blended — continues to offer the most balanced speed-to-intelligence ratio in the league for developers who need throughput without the premium model bill. 7

The speed race between Flash and Grok has been the storyline of the season. Day 11 was the apex — both cleared 206 t/s in a dead heat. Days 12 and 13 look like a controlled pullback rather than an injury; both models are still running well above their season-opening pace (183 and 145 t/s respectively on Day 1).

65 — Claude Fable 5's AI Index score, four points above any model posted in the first 12 game days of this season. The 12-day ceiling hold at 61 is over.

#AILeague