Tonight, the Largest Open-Weight AI Model in History Goes Live. It’s 1.4 Terabytes. It Beats Models Charging Real Money. Here’s What Actually Ships and What It Costs You: Nothing.

This week gave the industry two genuinely different answers to the same question: what does “the best available AI” actually cost right now? One answer arrived Thursday. The other arrives tonight.

Claude Opus 5 is Anthropic’s flagship model, launched July 24, 2026, priced at $5 input and $25 output per million tokens in standard mode — the same as its predecessor Opus 4.8, and half of Fable 5’s input price. It has a 1-million-token context window, a low/medium/high effort toggle, and scored 43.3% on FrontierBench v0.1. A fast mode runs at $10 and $50.

Moonshot AI’s Kimi K3 open weights release at 00:00 UTC on July 27, 2026 — tonight, in US time zones. The full weights are roughly 1.4 terabytes using MXFP4 quantization, making the 2.8-trillion-parameter model the largest open-weight release in history.

Anthropic’s flagship: genuinely capable, meaningfully cheaper than its own predecessor’s premium tier, and still metered per token. Kimi K3: larger than anything a closed lab has released this year, and once the weights land tonight, free to anyone with the hardware to run it.

Why Both of These Are Genuinely Good News, for Different Reasons

Opus 5 launching at the same price as Opus 4.8, rather than a premium markup, is worth noticing on its own — in a year where several launches have arrived with sharp introductory-pricing cliffs attached, holding a flagship at the previous generation’s price is a real signal that competitive pressure from open alternatives is already shaping how closed labs price their best models.

Kimi K3 going fully open at 2.8 trillion parameters is a different kind of milestone entirely — not because of the price (free), but because of the scale. This isn’t a smaller, distilled open model competing on efficiency. It’s the largest open-weight release that has ever existed, at a size that used to be reserved exclusively for the most heavily funded closed labs.

The Honest Caveat Neither Announcement Advertises

A 1.4-terabyte model isn’t something you casually run on a laptop. Self-hosting Kimi K3 at full scale requires serious dedicated hardware — the kind of infrastructure most individual developers and small teams don’t have sitting around, even with the weights being free. “Free” here means the per-token licensing cost is zero, not that the total cost of actually running it is zero.

That distinction matters for the comparison you should actually run this week: measure your real workloads against Opus 5, Kimi K3 (accounting for the infrastructure cost of self-hosting, not just the token price), and whatever you’re currently using — total cost, not just headline pricing.

Why This Week Specifically Deserves Your Attention

Combined with DeepSeek V4’s stable release landing just three days earlier on July 24, this closes out the single largest concentration of major open-weight releases the industry has seen — DeepSeek V4, Kimi K3’s full weights, and Google’s own Gemma 4 family, all within the same handful of days, alongside a closed flagship launch holding its price steady rather than increasing it.

That’s not a coincidence of timing. It’s the clearest evidence yet that the pricing pressure from open models is now directly shaping how closed labs launch their most important products — not in theory, but in the actual number on Anthropic’s pricing page this week.

Run the Real Comparison Tonight, Not Just the Spec Sheet Comparison

Whichever way this lands for your specific workload, the useful move isn’t picking a side between “open” and “closed” as a philosophy. It’s actually testing both against what you need, this week, while the comparison is more concrete than it’s been all year.

**aiexpo.app** — 1,600-plus tools, 70-plus categories, over 1,080 completely free, updated every single day — is where that honest comparison happens.

→ Compare Claude Opus 5 against every other frontier model, current pricing included: aiexpo.app/pages/category?cat=AI+Chatbots

→ Kimi K3, DeepSeek V4, Gemma 4, and every other open-weight tool, reviewed honestly: aiexpo.app/pages/tools

→ AI Coding Tools built on this week’s new models: aiexpo.app/pages/category?cat=AI+Coding+Tools

→ Track every release from this week and every week going forward: aiexpo.app/pages/new-tools

Anthropic held its flagship’s price steady this week instead of raising it. Kimi K3 ships tonight as the largest open-weight model that’s ever existed, for free. Both of those facts are direct evidence of the same underlying force: the pressure from open, free alternatives is no longer a future threat to closed labs’ pricing power. It’s already showing up in this week’s actual invoices.

Are you planning to actually test Kimi K3 tonight, or does the infrastructure requirement put it out of reach for your setup? Genuinely curious how many people can realistically self-host something at this scale.

🏠 aiexpo.app · 💬 Chatbots · 🔍 Free Tools · 💻 Coding Tools · 🆕 New Tools