July 12, 2026

Fable 5 Leaves Claude Subscriptions Tonight at $10/$50, and Four More Pricing Cliffs Are Coming

Today is the last day Fable 5 is included in Pro, Max, Team, and premium Enterprise plans. No third extension came. From Monday, every Fable token draws from usage credits at double Opus 4.8's rate, the most expensive model Anthropic sells. Gemini 3.5 Pro, DeepSeek V4, the OpenAI July 23 deprecation wave, and Sonnet 5's September price jump all land in the next 11 weeks.

Fable 5's included access ends tonight, and there is no third extension

The deadline has moved twice already. Fable 5 launched June 9 included in paid plans, was due to leave June 23, got pulled entirely on June 12 under a US export-control order, came back July 1 with a 50% weekly cap through July 7, and was extended to July 12 hours before that cutoff after subscriber backlash. Today, Sunday July 12, is the last day. At 11:59:59 PM Pacific tonight the promotional window closes for good, and Anthropic has not signaled another reprieve.

From Monday July 13, selecting Fable 5 in Claude, Claude Code, or the API draws from a separate prepaid usage-credit balance, not your Pro, Max, or Team weekly allowance. The rate is $10 per million input tokens and $50 per million output tokens, confirmed on Anthropic's official pricing page. That is double Claude Opus 4.8 ($5/$25) and the highest per-token price on Anthropic's entire rate card. Cache writes run $12.50 (5-minute) or $20 (1-hour) per million, cache hits and refreshes are $1, and batch processing drops it to $5/$25.

What to do before Monday: enable usage credits under Claude Settings and attach a payment method with a spending cap, or route Fable 5 workloads to Opus 4.8 or Sonnet 5, both of which stay inside your subscription limits unchanged. Standard seat-based Enterprise was already credits-only throughout this window, so nothing changes for those seats.

Two pieces of fine print that affect the bill. First, Fable 5's safety classifiers automatically reroute a small share of high-risk cybersecurity and biology requests to Opus 4.8, and Anthropic says you are not charged Fable prices for those rerouted requests. Second, all Mythos-class traffic carries a mandatory 30-day data retention for safety monitoring, so any workload that assumed zero retention needs a policy check. Mythos 5, the same underlying model with cyber safeguards lifted for vetted Project Glasswing partners, sits at the same $10/$50 but has no public sign-up path.

Anthropic has said it intends to restore Fable 5 as a standard subscription feature "when capacity allows." The deadline has moved twice in five weeks, so treat the Monday cutoff as the current policy, not necessarily the final one, but do not budget on another extension.

The countdown ahead: four pricing cliffs in 11 weeks

The Fable cliff is not the only deadline landing on developer budgets. Four more are now inside an 11-week window, and three of them hit in the next 12 days.

Gemini 3.5 Pro, target July 17 (5 days out, unconfirmed). Google has not confirmed the date. It traces to leaked server identifiers and prediction-market odds, not an official post. The I/O announcement on May 19 said only that Pro was "used internally" and would roll out "next month." Reports expect roughly $15/$60 per million tokens (about 10x Gemini 3.5 Flash) and a 2-million-token context window with a Deep Think reasoning mode, but Google has published no pricing, no benchmarks, and no model card. Watch the Gemini API pricing page and aistudio.google.com for the model card. Until it ships, Gemini 3.5 Flash at $1.50/$9 remains Google's GA frontier API model, with a free tier.

DeepSeek V4 official launch, mid-July (no exact date). When it lands, peak-hour pricing doubles API rates during 9:00-12:00 and 14:00-18:00 Beijing time, which is 1:00-4:00 AM and 6:00-10:00 AM UTC. Off-peak rates stay at the current V4 prices: V4 Pro at $0.435/$0.87 per million, V4 Flash at $0.14/$0.28. Cache-hit input is far cheaper, $0.003625 per million for Pro, which is how DeepSeek's automatic caching competes with formal batch discounts elsewhere. This is the first time-of-day surge pricing on a frontier AI API. Separately, the legacy deepseek-chat and deepseek-reasoner endpoints retire July 24 at 15:59 UTC (12 days out), routing to v4-flash until then.

OpenAI deprecation wave, July 23 (11 days out). The official deprecations page retires gpt-5-chat-latest, gpt-5.1-chat-latest, five Codex variants (gpt-5-codex through gpt-5.2-codex), computer-use-preview, o3-deep-research, o4-mini-deep-research, and the gpt-4o search-preview models. Replacements route to gpt-5.5, gpt-5.4-mini, or gpt-5.5-pro. The base gpt-5.4 model is not on this list despite a blog claim that circulated last week. Audit any hardcoded model IDs in gateway configs, Cursor, or Claude Code overrides before the 23rd, because most gateways do not auto-redirect retired names.

Sonnet 5 intro pricing ends September 1 (7 weeks out). The official pricing page confirms Sonnet 5 moves from $2/$10 to $3/$15 on September 1. The newer tokenizer, shared with Opus 4.7 and later, Fable 5, and Mythos 5, produces roughly 30% more tokens for the same text. So the effective increase is closer to $3.90/$19.50 per equivalent million tokens, which pushes Sonnet 5 above Sonnet 4.6's $3/$15 once the tokenizer cost is included.

Current prices: what every comparable model costs today

Prices verified against each vendor's official pricing or docs page on July 12 2026. All figures are per million tokens, input then output, unless noted.

• Claude Fable 5 at $10/$50, 1M context, 128K max output. Batch $5/$25. The most expensive model Anthropic prices. Credits-only from July 13.
• Claude Opus 4.8 at $5/$25, 1M context. Batch $2.50/$12.50. The standard Anthropic flagship, still included in subscriptions.
• Claude Sonnet 5 at $2/$10 through August 31, then $3/$15 on September 1. 1M context. New tokenizer adds roughly 30% more tokens.
• GPT-5.6 Sol at $5/$30, 1.05M context, 128K max output. OpenAI's flagship. Cache reads 90% off, 30-minute minimum cache life.
• GPT-5.6 Terra at $2.50/$15, 1.05M context. GPT-5.5-class performance at half the flagship price.
• GPT-5.6 Luna at $1/$6, 1.05M context. The cheapest OpenAI tier. Sol Fast adds a 2.5x premium at $12.50/$75 for 750 tokens per second.
• Gemini 3.5 Flash at $1.50/$9, 1M context, GA. Free tier available. Batch $0.75/$4.50. Priority tier $2.70/$16.20. Gemini 3.5 Pro is not yet GA.
• Grok 4.5 at $2/$6, 500K context (half of grok-4.3's 1M). Not on the batch-discount list. Free for a limited time in Grok Build and Cursor. EU access expected mid-July.
• Muse Spark 1.1 at $1.25/$4.25, 1M context, $0.15 cached input. Meta's first paid model. US-only waitlist, $20 free credits, not on OpenRouter. Thinking tokens billed at the output rate.
• DeepSeek V4 Pro at $0.435/$0.87 off-peak, 1M context, cache-hit input $0.003625. Doubles to $0.87/$1.74 during peak hours once the official V4 launch lands mid-July.
• Mistral Large 3 at $0.50/$1.50, Apache 2.0 open weights, 675B/41B-active MoE. A larger frontier open-weight MoE is in early access with no GA date or price yet.

That is the state of the market on a Sunday where the biggest move is a billing cliff, not a new model. The next two weeks will tell whether Google makes Gemini 3.5 Pro's July 17 target real, whether DeepSeek finally pins a date on V4, and whether any of these cliffs move again. Check the official pages above before you budget.