AI League — Game Day 3: DeepSeek Promo Clock Hits Zero, Claude Holds the Title

Claude Opus 4.8 keeps the #1 jersey. DeepSeek's 75% discount just expired at 15:59 UTC. Grok moves faster than anyone in the top 10. The stats don't lie. #AILeague

Live data from Artificial Analysis Intelligence Index v4.0 — 370+ models measured, updated every 72 hours.

The board is locked. No movement at the top this cycle — which itself is a story worth reading: Claude's margin over GPT-5.5 xhigh is now one point, the tightest gap this series.

Every stat on the board updated tonight. Here's the play that mattered.

The biggest play of Game Day 3 happened off the court. DeepSeek's promotional 75% discount — which slashed the V4 Pro output price from $3.48/M to $0.87/M — expired at 15:59 UTC today, May 31.1

Here's the twist the DeepSeek team buried in the footnotes: the promotion price is now permanent. According to the official pricing documentation, after the promo ends, DeepSeek V4 Pro pricing is officially adjusted to one-quarter of the original launch price — equal to the promotional rate. The "deal" becomes the list price.1

That means:

For context: GPT-5.5 xhigh output costs $30.00/M. DeepSeek's now-permanent output rate is 34× cheaper. At AI Index 52 vs. GPT's 60, the dark horse from Shenzhen is making a value argument no other team can match.2

Three teams are posting elite movement numbers this cycle — all confirmed via Artificial Analysis live performance benchmarks.2

Grok 4.3 (high) at 145.2 t/s is posting numbers no competing flagship reasoning model comes close to. xAI's squad has one stat that's genuinely hard to argue with: at $0.64/M blended with AI Index 53, it's the best intelligence-per-dollar-per-second ratio in the top 10.2

The post-DeepSeek-promo era begins tonight. Here's what the pricing grid looks like as of close of play, May 31.

GPT-5.5 xhigh holds the league's most expensive output price at $30.00/M — six times Gemini 3.1 Pro Preview and 34 times DeepSeek. OpenAI is betting the market will pay premium for that extra index point over Claude. Whether that 1-point edge at $30.00/M output is worth it compared to Grok's 53 at $2.50/M output is the pricing war question the league is asking.2

Kimi K2.6 remains the open-weights team to watch. At AI Index 54, it sits between Gemini 3.5 Flash (55) and Grok 4.3 (53) — above every fully open model in the league and priced at $0.70/M blended.2 Speed at 44.0 t/s is the weak spot; Kimi needs a pace upgrade to compete with Grok's 145 t/s.

MiMo-V2.5-Pro (Xiaomi) ties Kimi at AI Index 54 but with a blended price of $0.18/M — matching DeepSeek's permanent rate. The Xiaomi squad isn't on the mainstream radar yet, but that index-to-price ratio is one of the sharpest on the board.2

Meta's Llama 4 Scout is worth flagging not for intelligence (AI Index 14 by max-effort standard) but for context: its 10M-token window remains the widest in the league, and at $0.22/M blended with 104 t/s, it's the open-source community's speed-and-scale bargain.3

Meta's open-source squad still holds the context moat unchallenged — 10M tokens, 10× the next tier.3

Claude holds the title at AI Index 61 for a third consecutive game day — but the margin over GPT-5.5 xhigh is just 1 point, and at nearly 6× the output price, the value-for-intelligence question doesn't resolve in Anthropic's favor.2 Grok keeps posting 145 t/s while charging $2.50/M output — the closest thing in this league to a speed-value power play. And DeepSeek just locked in $0.18/M blended as a permanent rate, not a promo — a pricing floor the rest of the market will need to answer.

The clock hit zero on one era. The next game starts now.

#AILeague · Data: Artificial Analysis · DeepSeek API Pricing