The Shakeup
Today marks the first time a single model has simultaneously seized the #1 spot in both Code Generation and Text Generation: DeepSeek V4 Pro Max dropped today and immediately dethroned incumbents in two major categories. Meanwhile, MiniMax H3 quietly took the Image-to-Video crown, and a wave of 10 new models crashed into the top-10 rankings across five categories — making this one of the most turbulent days on the leaderboards in weeks.
What Changed Today
- 🥇 New #1 in Code Generation:
deepseek-v4-pro-max-20260813overtookclaude-opus-5-max - 🥇 New #1 in Text Generation:
deepseek-v4-pro-max-20260813overtookinkling-small - 🥇 New #1 in Image-to-Video:
minimax-h3overtookdreamina-seedance-2.0-720p - 🆕 Code Generation:
grok-4.6-highentered the top 10 at ELO 1630;deepseek-v4-pro-max-20260813entered at ELO 1607 - 🆕 Text Generation:
claude-opus-4-7-high(ELO 1502) andclaude-opus-4-6-high(ELO 1505) entered the top 10 alongside DeepSeek (ELO 1466) - 🆕 Text-to-Video:
flux-3-videoentered the top 10 at ELO 1496 - 🆕 Image-to-Video:
flux-3-video-20260811entered the top 10 at ELO 1453 - 🆕 Vision:
claude-opus-4-7-high(ELO 1301),claude-opus-4-6-high(ELO 1300), andgemini-3.6-flash-high(ELO 1295) all entered the top 10
Code Generation: DeepSeek Takes the Throne
DeepSeek V4 Pro Max (deepseek-v4-pro-max-20260813) launched today and immediately claimed the #1 spot in Code Generation with an ELO of 1607, dethroning claude-opus-5-max. This is a significant moment — DeepSeek has been climbing the coding ranks for months, but this is the first time a DeepSeek model has held the outright #1 position in code. For developers building coding agents or code-generation pipelines, this is worth benchmarking against your current Anthropic or OpenAI setup immediately.
Notably, Grok 4.6 High (grok-4.6-high) also entered the top 10 today at an even higher raw ELO of 1630, suggesting xAI’s latest coding model is extremely competitive. It’s unclear why it isn’t ranked #1 — likely due to vote count thresholds or confidence intervals — but keep a close eye on this one. Pricing for both models hasn’t been confirmed in our reference data, so check the respective API dashboards before committing to production workloads.
FREE GUIDE
Stop Writing Design Specs by Hand
Get the free visual guide: how AI tools generate GAMP 5 documentation directly from your PLC and DCS exports. Used by Life Sciences engineers who are done doing it manually.
No spam. Unsubscribe anytime.
Text Generation: DeepSeek Doubles Down
DeepSeek V4 Pro Max didn’t stop at code — it also took the #1 position in Text Generation, overtaking inkling-small. Its text ELO of 1466 is notably lower than the previous leader tier (the reference doc shows claude-opus-4-6 at 1504), which suggests the leaderboard has seen significant reshuffling since our last reference snapshot. The dual crown in code and text makes deepseek-v4-pro-max-20260813 an interesting “one model to rule them all” candidate for teams that want to consolidate their API providers.
Also entering the text top 10 today: Claude Opus 4.7 High (claude-opus-4-7-high, ELO 1502) and Claude Opus 4.6 High (claude-opus-4-6-high, ELO 1505). These “high” variants from Anthropic appear to be compute-upgraded versions of existing Opus models, and both score extremely well — developers already on Anthropic’s API should check if these are available under their current plan. Given Anthropic’s typical pricing structure around $5/$25 per MTok for Opus-class models, expect similar rates.
Image-to-Video: MiniMax H3 Takes #1
MiniMax H3 (minimax-h3) has overtaken dreamina-seedance-2.0-720p to become the new leader in Image-to-Video generation. MiniMax has been a rising force in video generation — their M2.5 model already sits in the code top 10 — and H3 represents their most capable video model yet. For developers building image-animation features or visual content pipelines, MiniMax’s API is worth evaluating, particularly if you’ve been locked into Google’s Veo ecosystem or xAI’s Grok Imagine Video.
Also new in Image-to-Video: Flux 3 Video (flux-3-video-20260811) entered at ELO 1453, which is a strong debut. Black Forest Labs continues to expand from their image-generation dominance into video territory, and an ELO of 1453 puts Flux 3 Video comfortably in the top tier. This model also appeared in the Text-to-Video top 10 as flux-3-video (ELO 1496), making it a versatile option for teams that need both text-to-video and image-to-video from a single provider.
Vision: Anthropic and Google Push New Variants
The Vision leaderboard saw three new entrants today: Claude Opus 4.7 High (ELO 1301), Claude Opus 4.6 High (ELO 1300), and Gemini 3.6 Flash High (ELO 1295). All three score above the current #1 in our reference data (gemini-3-pro at 1288), which means the vision leaderboard likely has a new leader among these models. For vision-heavy workloads — document understanding, image analysis, multimodal agents — these “high” compute tiers from both Anthropic and Google are delivering meaningful quality gains over their standard counterparts.
Text-to-Video: Flux Enters the Chat
Flux 3 Video (flux-3-video) debuted in the Text-to-Video top 10 at ELO 1496 — which, if accurate, would make it the new #1 in the category by a significant margin over the previous leader veo-3.1-audio-1080p (ELO 1392). This is a massive leap and worth watching closely as vote counts accumulate. Black Forest Labs pricing and API access details aren’t yet in our reference data, so early adopters should check their developer portal directly.
Current Leaders at a Glance
| Category | #1 Model | Provider | ELO Score |
|---|---|---|---|
| Code Generation | deepseek-v4-pro-max-20260813 | DeepSeek | 1607 |
| Text Generation | deepseek-v4-pro-max-20260813 | DeepSeek | 1466 |
| Image-to-Video | minimax-h3 | MiniMax | TBD |
| Text-to-Video | veo-3.1-audio-1080p | 1392 | |
| Vision | gemini-3-pro | 1288 |
Note: Vision and Text-to-Video may have new de facto leaders based on today’s new entrants scoring above incumbents. We’ll confirm once vote counts stabilize.
So What?
Today’s biggest takeaway is simple: DeepSeek is now a top-tier contender you can’t ignore. If you’re building coding agents, AI-assisted development tools, or text-generation features and you’ve been defaulting to Anthropic or OpenAI, it’s time to run deepseek-v4-pro-max-20260813 through your eval suite — a model that tops both code and text simultaneously deserves a serious look, especially if DeepSeek’s pricing undercuts the incumbents (as it historically has). On the video side, the market is fragmenting fast: MiniMax, Flux, Google, and xAI are all trading blows across image-to-video and text-to-video, which means competition is driving quality up and prices down — great news for builders. The practical move today is to add DeepSeek V4 Pro Max to your model routing options, evaluate flux-3-video if you’re doing any video generation work, and keep an eye on the “high” compute variants from Anthropic and Google — they’re quietly outscoring their standard siblings across text, code, and vision, suggesting that paying for extra inference compute is becoming a reliable quality lever worth building into your tiered product offerings.
