Fable 5 Permanent Split Goes Live July 20: Max Keeps It, Pro Pays $10/$50

The five-week Fable 5 deadline drama ended on schedule at midnight Pacific with no fourth extension. DeepSeek V4 general availability, reported by Chinese press for "as early as Monday," had not officially landed as of this morning. The Claude Code 50% usage boost expired overnight, compounding the cut for Max subscribers.

Fable 5's permanent tier split took effect today, on schedule

The cliff this feed has tracked through three wire-moves finally resolved, and it resolved on time. At midnight Pacific on July 20, Anthropic put the permanent two-tier access structure for Claude Fable 5 into effect, exactly as the July 18 @claudeai announcement specified, with no last-minute fourth extension. TechTimes confirmed the cutover (July 20, 6:20am EDT), and Android Authority verified that Fable 5 is "now officially included with Max and Team Premium subscriptions."

What landed, plan by plan:

• Max and Team Premium keep Fable 5 bundled into the subscription at no per-token charge, capped at 50% of standard weekly usage limits, indefinitely. This is the permanence the reader has been waiting on since the June 23 original deadline.
• Pro and Team Standard lose bundled access entirely. They receive a one-time $100 usage credit, after which Fable 5 bills at $10 per million input tokens and $50 per million output tokens, the steepest rate on Anthropic's generally available price card and double Opus 4.8's $5/$25.

The arithmetic on that $100 credit is the part worth internalizing. At $50 per million output tokens, a single intensive agentic coding or document-analysis session that emits 2 million output tokens, a moderate load for the multi-file autonomous work Fable 5 was built for, consumes the entire credit. Pro subscribers who leaned on Fable 5 for daily work now face a binary choice: upgrade to Max ($100 or $200 per month, versus Pro's $20), route heavy work to a cheaper model, or accept metered billing that scales with output volume.

There is a compound effect on the Max side too, and it hits the same overnight. The 50% boost to Claude Code weekly rate limits, in place since May 13 and extended alongside each Fable 5 deadline, also expired with this cutover. Standard limits shrink first, then the 50% Fable cap applies against that reduced baseline. A Max subscriber who consumed heavily during the promotional window will feel compression on both ends: a lower Code ceiling and a halved Fable allowance measured against it.

Why Anthropic held the line this time instead of extending again is competitive, not just operational. GPT-5.6 Sol reached general availability July 9 at $5/$30, roughly half Fable 5's input cost and 60% of its output rate, and scores within a point of Fable 5 on the Artificial Analysis Intelligence Index, a statistical tie on general capability while completing equivalent tasks at about one-third Fable 5's per-task cost. A Max plan offering no access to Anthropic's own flagship, while a competitor sells near-equivalent performance for less, was a retention problem. The split preserves Fable 5 as a differentiated feature for the highest-revenue consumer tiers and prices Pro users toward Sonnet 5 or an upgrade.

The full deadline sequence, for the record: June 23 original, pulled June 12 under the Commerce export control, restored July 1, extended to July 7, extended to July 12, extended to July 19, then resolved July 18 as a permanent split effective July 20. Three wire-moves, then a policy change rather than a cliff. API builders are unaffected: Fable 5 on the API stays $10/$50, with cache hits at $1, 5-minute cache writes at $12.50, and 1-hour cache writes at $20, per the Anthropic pricing page.

DeepSeek V4 GA was reported for today, but had not officially landed as of this morning

Yesterday's issue flagged Chinese tech press reporting that DeepSeek V4 full general availability could come "as early as Monday, July 20." Monday is here, and the honest status is: it had not officially landed as of 6:00 AM Pacific.

The signal to watch is DeepSeek's own API Change Log, and it has not moved. The official changelog still lists "2026-04-24: DeepSeek-V4" as its most recent entry, the April preview launch. There is no July news post, no GA announcement, and no new model-card entry. The pricing page still shows the baseline off-peak rates that have been live since the preview: V4 Pro at $0.435 input / $0.87 output per million tokens, V4 Flash at $0.14 / $0.28, both with a 1-million-token context and 384K max output. No peak-pricing row has appeared.

What is out there is reporting, not an official launch. 36kr wrote July 19 that "the official release of DeepSeek V4 may happen as early as tomorrow, and no later than the next few days," citing grayscale-test access already circulating to some users. The Standard HK matched that "as early as Monday" framing, adding that performance approaches Opus 4.8. These are credible Chinese outlets, but they are reporting on a window, not announcing a ship. A few weak domains have already published "V4 GA is here" roundups dated July 19, the same content-mill pattern that produced the false "Gemini 3.5 Pro launched July 17" posts this feed debunked last week. Treat any per-token GA claim as unconfirmed unless a new entry appears on the DeepSeek changelog.

What is hard and confirmed is the deadline four days out. The legacy deepseek-chat and deepseek-reasoner model names retire July 24 at 15:59 UTC, routing to deepseek-v4-flash non-thinking and thinking modes until then. Anyone still calling the old names has four days to rename. When GA does land, the peak-valley pricing announced June 30 via TechNode goes live: peak hours of 9:00-12:00 and 14:00-18:00 Beijing time (1:00-4:00 AM and 6:00-10:00 AM UTC) bill at double the off-peak rate, meaning V4 Pro peaks at $0.87 / $1.74 and V4 Flash at $0.28 / $0.56. That makes DeepSeek the first frontier API to charge by time of day. For US and European workloads those peak windows fall overnight, so the practical hit is small outside China.

Tracking

• OpenAI July 23 deprecation wave, 3 days out. Fifteen listed model entries shut down July 23 per the OpenAI deprecations page: gpt-5-chat-latest, gpt-5.1-chat-latest, five Codex variants (gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini, gpt-5.2-codex), computer-use-preview, two deep-research entries, two search previews, and audio, TTS, and realtime-mini snapshots. The recommended successor for most is gpt-5.5. Note that gpt-5.4 base is NOT on this list, despite some aggregator claims to the contrary.

• Kimi K3 open weights, July 27 (7 days). Moonshot's 2.8-trillion-parameter model, the largest open-weight ever shipped, has been API-only since its July 16 launch at $3/$15 per million tokens with a flat 1-million-token context. Weights are scheduled for July 27 under a modified-MIT license, which would reshuffle the open-weight leaderboard (K3 2.8T ahead of Qwen 3.8 2.4T, DeepSeek V4 Pro 1.6T, and Inkling 975B).

• Gemini 3.5 Pro, still missing. Google's frontier target has now missed June, July 17, and beyond. The Google API changelog still lists only gemini-3.5-flash as generally available, and the pricing page still has no 3.5 Pro row, only Flash at $1.50/$9, Flash-Lite at $0.25/$1.50, and 3.1-pro-preview at $2/$12. False "launched July 17" posts on weak domains are still circulating. Treat any Gemini 3.5 Pro "GA" claim as false unless gemini-3.5-pro appears in the public API docs.

• Sonnet 5 price step, September 1. Intro pricing of $2/$10 runs through August 31, then steps to $3/$15. The new tokenizer emits roughly 30% more tokens than the old one, so the effective rate from September 1 is about $3.90/$19.50, more than Sonnet 4.6's $3/$15. Pro subscribers pushed off Fable 5 today should note Sonnet 5's intro rate is the cheaper redirect while it lasts.

• Mistral frontier MoE, still early access. CEO Arthur Mensch confirmed a "fat but sparse" open-weight Mixture-of-Experts model entered early access in July with research, government, and industry partners. No name, no parameter count, no benchmarks, no price yet. General availability is expected later this summer.

• Qwen 3.8-Max-Preview, launched yesterday. Alibaba's 2.4-trillion-parameter preview is live on the Token Plan, credits-only with no per-token API price yet, claiming "second only to Fable 5" on internal evals with no independent benchmarks. Open weights are "soon." The Qwen developer pricing page lists qwen3.7-max at $2.50/$7.50 but still has no qwen3.8 row.

Current prices, July 20

Per million input and output tokens, linked to each vendor's official pricing page. All pages verified live this week except where noted.

• GPT-5.6 Sol - $5 / $30, 1.05M context. developers.openai.com (verified Jul 18). Sol Fast $12.50 / $75 via Cerebras.
• GPT-5.6 Terra - $2.50 / $15, 1.05M context. developers.openai.com (verified Jul 18).
• GPT-5.6 Luna - $1 / $6, 1.05M context. developers.openai.com (verified Jul 18).
• Claude Fable 5 - $10 / $50 (API and now the Pro/Team Standard usage-credit rate). platform.claude.com (verified Jul 18).
• Claude Opus 4.8 - $5 / $25. platform.claude.com (verified Jul 18).
• Claude Sonnet 5 - $2 / $10 intro through Aug 31, then $3 / $15 (effective ~$3.90/$19.50 after the tokenizer tax). platform.claude.com (verified Jul 18).
• Gemini 3.5 Flash - $1.50 / $9, 1M context (still Google's live frontier; 3.5 Pro not launched). ai.google.dev (verified Jul 18).
• Grok 4.5 - $2 / $6, 500K context. docs.x.ai (verified Jul 18).
• DeepSeek V4 Pro - $0.435 / $0.87 off-peak, doubles at peak once GA lands. api-docs.deepseek.com (verified Jul 20, still baseline).
• DeepSeek V4 Flash - $0.14 / $0.28 off-peak. api-docs.deepseek.com (verified Jul 20, still baseline).
• Kimi K3 - $3 / $15, flat across 1M context, cache hit $0.30. VentureBeat (no official Moonshot per-token page crawled).
• Meta Muse Spark 1.1 - $1.25 / $4.25, $0.15 cached, 1M context. dev.meta.ai (verified Jul 18).
• Mistral Large 3 - $0.50 / $1.50, Apache 2.0, 675B/41B-active. mistral.ai (verified Jul 18).
• Qwen 3.7-Max - $2.50 / $7.50. docs.qwencloud.com (verified Jul 19; Qwen 3.8 is credits-only, no per-token API yet).

The frontier output-price ladder today runs from DeepSeek V4 Flash at $0.28 to Fable 5 at $50, a 178x spread inside one market. Where a model sits on that ladder now correlates more with context window, safety retention overhead, and time-of-day billing than with raw benchmark scores, which is the structural shift worth watching as DeepSeek's peak pricing and Mistral's open-weight frontier both approach.