May 29, 2026 · 11:27 AM
AI League — Season Opening Night: The Official Stats Panel, Week 1
Claude Opus 4.8 tops the board (AI Index: 61). DeepSeek V4 Pro cuts output price 75% to $0.87/M. Gemini 3.5 Flash hits 207 t/s. Full post-game stats panel. #AILeague
Presented by AI League Daily. Data sourced from Artificial Analysis, official API pricing pages, and OpenRouter usage rankings.
Claude Opus 4.8 takes the #1 slot on the Artificial Analysis Intelligence Index. GPT-5.5 drops to second. Gemini 3.5 Flash sets a season-high speed record at 207 t/s. And DeepSeek just slashed V4 Pro output pricing by 75% — the biggest in-season price cut any team has posted this year.
🏆 Overall standings — Intelligence Index (Artificial Analysis, live)
| Rank | Team | Model | AI Index | Δ | Price/1M blend |
|---|---|---|---|---|---|
| 🥇 1 | Anthropic (Claude) | Claude Opus 4.8 (Reasoning Max) | 61 | ↑ NEW | $4.10 |
| 🥈 2 | OpenAI (GPT) | GPT-5.5 (xhigh) | 60 | — | $4.35 |
| 🥉 3 | OpenAI (GPT) | GPT-5.5 (high) | 59 | — | $4.35 |
| 4 | Anthropic (Claude) | Claude Opus 4.7 (Reasoning Max) | 57 | ↓ −1 | $4.10 |
| 5 | Google (Gemini) | Gemini 3.1 Pro Preview | 57 | ↑ NEW | $1.74 |
| 13 | xAI (Grok) | Grok 4.3 (high) | 53 | — | $0.64 |
| 17 | DeepSeek | DeepSeek V4 Pro (Reasoning Max) | 52 | — | $0.18 |
| 18 | Meta (Llama) | Muse Spark | 52 | ↑ NEW | — |
Artificial Analysis Intelligence Index (0–100 composite score, live 72-hr rolling window)
1
🔥 Game of the night: Claude vs. GPT in the scoring race
Claude Opus 4.8 posts an Intelligence Index of 61 to GPT-5.5's 60 — a one-point margin that is statistically tight on Artificial Analysis's composite methodology, but Claude has now held the top position across successive measurements. GPT-5.5 remains the higher-volume franchise: it fields three active pricing tiers (xhigh, high, medium) priced at a flat $4.35/1M blend compared to Claude Opus 4.8's $4.10/1M, a 6% pricing edge for Anthropic at equal performance.
Meanwhile, the legacy GPT-4o line (not on the current Artificial Analysis front page) has been effectively benched in favor of the GPT-5 series. OpenAI's pricing page no longer lists standalone GPT-4o rates — the franchise is running on a GPT-5.4 / GPT-5.5 two-tier rotation.
23⚡ Speed panel — tokens per second (Artificial Analysis live)
This is where the scoreboard gets interesting. Intelligence and speed almost never correlate.
| Team | Model | Output speed (t/s) | TTFT (s) | Speed tier |
|---|---|---|---|---|
| Google (Gemini) | Gemini 3.5 Flash | 207 t/s | 16.2s | 🚀 Season-high |
| Google (Gemini) | Gemini 3.1 Flash-Lite | 268 t/s | 5.7s | 🚀 Fastest Gemini |
| Anthropic (Claude) | Claude Sonnet 4.6 (Reasoning Max) | 120 t/s | 110.7s | ⚡ Fast-reasoning |
| Anthropic (Claude) | Claude Haiku 4.5 | 97 t/s | 16.9s | ⚡ Budget burner |
| xAI (Grok) | Grok 4.3 (high) | 182 t/s | 12.5s | 🏃 Underrated pace |
| OpenAI (GPT) | GPT-5.5 (xhigh) | 40.7 t/s | 49.1s | 🐢 Deliberate |
| DeepSeek | DeepSeek V4 Pro (Reasoning Max) | 50 t/s | 1.8s | ⚡ Low TTFT |
Season stat: Gemini 3.5 Flash's 207 t/s is the fastest production speed among models in the AI Index top 10 score bracket (55+). Google may be outside the intelligence top 5, but they're running the fastest offense of any team in that tier.
Grok's quiet hustle: Grok 4.3 (high) at 182 t/s with 12.5s TTFT punches above its Intelligence Index rank of 53. If xAI can close the scoring gap, the speed advantage becomes a real weapon.
💰 Pricing war analysis — the cost-per-point breakdown
The real competition is in the value tier — Intelligence Index points per dollar of API cost.
| Team | Flagship model | Input $/1M | Output $/1M | AI Index | Points per $1 output |
|---|---|---|---|---|---|
| DeepSeek | V4 Pro (75% promo) | $0.435 | $0.87 | 52 | 59.8 |
| Gemini 3.5 Flash | $0.30* | $2.50* | 55 | 22.0 | |
| xAI | Grok 4.3 | $1.25 | $2.50 | 53 | 21.2 |
| Gemini 2.5 Pro | $1.25 | $10.00 | 35 | 3.5 | |
| Anthropic | Claude Sonnet 4.6 | $3.00 | $15.00 | 52 | 3.5 |
| OpenAI | GPT-5.5 | $5.00 | $30.00 | 60 | 2.0 |
| Anthropic | Claude Opus 4.8 | $5.00 | $25.00 | 61 | 2.4 |
Gemini 2.5 Pro/Flash pricing from Google AI Developer pricing page; DeepSeek pricing per official API docs (75% discount valid through May 31, 2026 UTC); Grok from xAI API page.
23456DeepSeek's fire sale
DeepSeek's V4 Pro is currently on a 75% promotional discount — output at $0.87/1M versus the post-promotion rate of $3.48/1M. With an Intelligence Index of 52, it delivers roughly 60 AI-index points per dollar of output, compared to GPT-5.5's 2.0. This promo expires May 31, 2026. If DeepSeek holds that line post-discount, it becomes a real threat to the mid-market. If it reverts to full price, the equation changes entirely.
The macro pattern: Google and DeepSeek are playing a volume game. Gemini 3.5 Flash at $0.30/$2.50 in/out and Gemini 2.5 Flash at $0.30/$2.50 are the lowest-priced capable models from any named team. Google's free tier (via Google AI Studio) makes it the only franchise offering zero-cost access to a competitive model.
📊 Context window depth chart
| Team | Model | Context window | Field advantage |
|---|---|---|---|
| Meta (Llama) | Llama 4 Scout | 10M tokens | Widest context in the league |
| xAI (Grok) | Grok 4.20 / 4.1 Fast | 1M tokens | Second-widest |
| Google (Gemini) | Gemini 1.5 Pro (May) | 1M tokens | Third-widest |
| Anthropic (Claude) | Claude Opus 4.8 | 1M tokens | Tied third |
| OpenAI (GPT) | GPT-5.5 | 922k tokens | Slight edge over prior 128k window |
| DeepSeek | DeepSeek V4 Pro/Flash | 1M tokens | Full million |
Meta's Llama 4 Scout at 10 million tokens is the long-haul freight runner of this league — no other model is within 10x of its context capacity. That's a specialized stat, but for enterprises running massive document pipelines it's a decisive edge.
1🆕 Challenger watch — challengers making noise this cycle
- Kimi K2.6 (Moonshot AI) — Intelligence Index 54, ranking it above DeepSeek V4 Pro and all named Grok variants. Priced at $0.70/1M blend. An unseeded challenger crashing the main draw.
- Mistral Medium 3.5 — AI Index 39, 152 t/s, priced at $2.10/1M blend. Solid mid-tier performance, especially for EU-regulated deployments.
- Cohere Command A+ — AI Index 37, 223 t/s output speed — the highest raw token throughput on the entire leaderboard among models with a price listed. Zero-cost pricing (free tier). Useful for latency-critical applications but the intelligence gap versus tier-1 models is wide.
🏁 Postgame summary
Claude Opus 4.8 holds the scoring record. GPT-5.5 stays in contention on depth and ecosystem. Google's Gemini family is running the fastest offense in the ability-tier. DeepSeek's discount is the price-war story of the cycle — $0.87/1M output won't last past May 31. Meta's Llama 4 Scout has a context advantage no other franchise can touch. Grok is faster than its rankings suggest.
The teams that should be watching? Anyone who isn't Google or DeepSeek in the budget tier is currently fighting for value at a significant per-token disadvantage.
Data as of May 29, 2026 — live metrics from Artificial Analysis (72-hr rolling), official pricing from each team's API documentation pages. Speed metrics reflect streaming output tokens per second under standard single-request conditions.
#AILeague
Related content
Picked from other channels by content similarity—find new creators to follow.
Article还没有赢家：2026 年 5 月大模型竞争全景
Claude Opus 4.7 在编程 Agent 领域领跑，但综合智能指数第一是 GPT-5.5。Gemini 3.1 Pro 以 40% 的价格交付 80% 的性能，Grok 4.3 靠实时知识打差异化。Anthropic 估值飙至 $1.2 万亿背后是 Claude Code 的企业渗透逻辑，而开源模型正在改变「赢」的定义本身。
Twitter AI 长文精选
ArticleAIL Player Card #004 — DeepSeek V4 Pro: The Value Engineer
95 OVR. VE. Open-source. 1.6T parameters. $3.48/M output vs $30 for GPT-5.5. Codeforces ELO 3206 — beats the incumbent by 38 points. DeepSeek Athletic just repriced the frontier. #AILeague
AIL·Player Card
ArticleAIL Player Card #009 — Gemini 3.5 Flash: The Agentic Sprinter
91 OVR. AS. Terminal-Bench 76.2%. SWE-Bench Pro 55.1%. 4× faster than frontier rivals. $1.50/M input. Google National finally fields a player built for the agentic era — faster, cheaper, and better at multi-agent loops than its own flagship. #AILeague
AIL·Player Card
Article四大模型同日登场，五角大楼清洗 AI 供应商 | 5月1日
GPT-5.5、Claude Opus 4.7、DeepSeek V4、Grok 4.3 同日亮相，闭源模型越来越贵、开源竞品越来越能打，剪刀差持续扩大；美国国防部重新划定 AI 供应商名单，Anthropic 因拒绝放松「自主武器」使用限制而被踢出；Meta 收购人形机器人公司 ARI 入局具身 AI，Musk v. Altman 庭审中 xAI 用 OpenAI 模型蒸馏 Grok 的事实被坐实；Salesforce 发布企业 Agent 运维平台，Netomi 完成 1.1 亿美元 C 轮。
AI 日报｜量子位风
ArticleAIL Player Card #013 — Claude Fable 5: The Mythos Striker
95 OVR. SF. SWE-Bench Pro 80.3% — 20 pts clear of GPT-5.5. FrontierCode Diamond 29.3%. Mythos-class. First Anthropic model you can actually buy. Anthropic FC just cleared their best player for the public pitch. #AILeague
AIL·Player Card
ArticleGPT-5.5 is out — and it closes the gap on reasoning while cutting token costs in half
OpenAI shipped GPT-5.5 on April 23, 2026 — its most capable model to date, with major benchmark gains over GPT-5.4 and per-token costs roughly half those of comparable frontier models. This article covers what changed, how it stacks up against Claude Opus 4.7, API pricing, and the competitive implications including DeepSeek V4's arrival the next day.
AI Model & Product Launch Alerts
Add more perspectives or context around this Post.