Week 30 Briefing: AI Pricing Updates: July 20-26, 2026

Anthropic ships Opus 5 at half Fable 5's price. Kimi K3 hits GPU capacity wall and pauses new subscriptions. Two big moves, one clear signal: the price war continues but infrastructure is the bottleneck.

💰 Your Wallet This Week

#1 Save $: Claude Opus 5 launched July 24 at $5/$25 per million tokens, half the price of Fable 5. Pro subscribers keep Opus 5 access. Max users get it at 50% of limits, same as Fable 5.
#2 Locked Out: Kimi K3 new subscriptions paused since July 19 due to GPU capacity limits. Existing subscribers are protected. No timeline for reopening. If you were on the fence, you are now stuck waiting.
#3 Heads Up: Anthropic’s price strategy signals a permanent shift: premium model quality at mid-tier pricing is now the baseline. Opus 5 costs the same as Opus 4.8 did, but approaches Fable 5 benchmarks. Expect pressure on OpenAI and Google to respond.

Price Change Summary

Tool	What Changed	Was	Now	Impact	Who It Affects
🧠 Anthropic Opus 5	New model launched. Available on Pro and Max plans. API at same price as Opus 4.8.	N/A	$5/M in + $25/M out	Cost Down	Claude Pro/Max/API users
🤖 Kimi K3	New subscriptions paused. GPU capacity maxed. Existing subs protected.	$19-199/mo	Paused	Unavailable	New Kimi users

🗃 Week 29: Fable 5 Pro cutoff, Meta Muse Spark 1.1, Grammarly rebrand · Week 28: DeepSeek V4 Pro, Perplexity Spaces, Grok Imagine

🧠

Anthropic Launches Claude Opus 5: Near Fable 5 Intelligence at Half the Price

Anthropic shipped Claude Opus 5 on July 24, and the pricing decision is the real story. The model costs $5 per million input tokens and $25 per million output tokens through the API, the same as Opus 4.8. The headlines say “half price” because those numbers are 50% less than Fable 5 at $10/$50. For subscription users, Opus 5 is available on both Pro ($20/mo) and Max ($100-200/mo) plans starting day one.

Benchmarks put Opus 5 within 0.5% of Fable 5 on SWE-bench Pro at max effort. On Frontier-Bench it scores 43.3% (state-of-the-art), placing it clearly ahead of GPT-5.6 Sol. The catch is a new effort toggle: low/medium/high settings let you trade speed for depth. High effort uses more compute and hits rate limits faster, but you control the tradeoff.

Anthropic also offers a Fast Mode at $10/$50 per million tokens (2.5x speed for 2x price). This mirrors OpenAI's responsive tier, giving developers a premium option for latency-sensitive applications.

The strategic message is clear: Anthropic is making premium model quality the new mid-tier baseline. Opus 5 costs the same as Opus 4.8 but approaches Fable 5 performance. That puts pressure on OpenAI to justify GPT-5.6 Sol’s $5/$30 pricing against a model that nearly matches it for slightly less on output costs.

✅ Who This Affects

Claude Pro and Max subscribers. Pro users keep Opus 5 access at standard limits. Max users get it at 50% of plan caps (same treatment as Fable 5). API developers should benchmark Opus 5 against Fable 5: if your workload doesn't need peak Fable 5 performance, switching to Opus 5 saves 50% on API costs. The effort toggle also lets you use low/medium modes for simple tasks and save high for complex ones, effectively a built-in cost optimizer.

🤖

Kimi K3 Subscriptions Paused as Demand Overwhelms GPU Capacity

Moonshot AI paused all new Kimi K3 subscriptions on July 19, less than a week after launch. Demand ran 6x above expectations, overwhelming the company’s GPU compute. Moonshot posted: “Kimi K3 has received far more love than we expected, and our GPUs are feeling it. To protect the experience of existing subscribers, we are temporarily pausing new subscriptions.”

The subscription tiers remain unchanged for existing users: Moderato at $19/mo (1x), Allegretto at $39/mo (5x), Allegro at $99/mo (15x), and Vivace at $199/mo (30x). All are protected during the pause. Moonshot says it will reopen new subscriptions in batches as capacity comes online.

One structural change is coming: Moonshot plans to split Kimi Membership (web, app, Kimi Work) from Kimi Code Membership (coding workflows). This gives the company a cleaner way to price and allocate compute between general users and developers. The split begins July 20, but existing subscribers won’t see pricing changes.

K3’s open weights, originally slated for July 27, are still on schedule as of this writing. If that release happens, developers will have a third path: run K3 locally via self-hosted infrastructure, bypassing both the subscription pause and the rate limits entirely.

⚠️ Who This Affects

Anyone who was waiting to try Kimi K3. New subscriptions are paused with no stated reopening date. If you held off during launch week to see benchmarks and reviews, you missed the window. Existing subscribers are protected. If you already have a Kimi subscription, you keep full access. Developers: watch for K3 open weights on July 27. Self-hosting could sidestep the capacity problem entirely if you have local compute.

🔍 Switch Watch

Three recommendations based on this week’s pricing changes.

Opportunity

Claude API Users: Test Opus 5 Against Your Fable 5 Workloads

If your application runs on Fable 5, benchmark it against Opus 5 at high effort. Many tasks will see near-identical output quality. Switching saves 50% per token. For latency-sensitive work, Opus 5 at medium effort with Fast Mode at $10/$50 still undercuts Fable 5 pricing while matching speed.

Action Required

Waiting on Kimi K3? You are Out of Luck Until Capacity Opens

No new subscriptions until Moonshot adds GPU capacity. No timeline given. If K3 is critical to your workflow, check OpenCode Go ($10/mo) which still offers K3 access under its own quota allocation. That is currently the only path for new users.

Info

Claude Pro Users: Your Fable 5 Access Ended July 20

Week 29 warned about this. It is now in effect. Pro subscribers get a one-time $100 credit for Fable 5 API usage. Max users keep Fable 5 at 50% of plan limits. If you need Fable 5 on a Pro budget, the credit buys roughly 2 million output tokens before you hit pay-per-use at $50/M out.