July 13, 2026

Fable 5 Didn't Leave Subscriptions: Extended to July 19, and GPT-5.6 Sol Gets 10% More Usage

The deadline that was supposed to land yesterday moved again, to July 19. It is the third time the Fable 5 cutoff has shifted, each one resolved at the wire. Meanwhile OpenAI landed inference optimizations for GPT-5.6 Sol that give subscribers roughly 10% more usage, and temporarily lifted the five-hour rate limit.

Fable 5's cliff moved to July 19, superseding yesterday's deadline

Yesterday's issue reported that Fable 5's included subscription access would end at 11:59:59 PM PT on July 12, with no further extension. It did not. Anthropic announced via @claudeai on X that plan-included Fable 5 access on all paid plans, plus Claude Code's 50% higher weekly rate limits, now run through July 19, 2026 at 11:59:59 PM PT. Anthropic's official Fable page and support documentation were updated to match.

The rules are unchanged. Eligible plans (Pro, Max, Team, and premium seats on seat-based Enterprise) can spend up to 50% of their weekly usage limits on Fable 5 at no extra cost. Fable 5 draws from the same weekly pool as every other Claude model and consumes it faster, so the 50% allotment is a ceiling within existing limits, not bonus capacity. Once that ceiling is hit, the choice is prepaid usage credits at $10/$50 per million tokens or switching to another model. Free users, standard Enterprise seats, usage-based Enterprise plans, and API traffic are excluded from the promotion.

The pattern is now the story. The cutoff has moved three times: from June 23 to July 7, from July 7 to July 12, and from July 12 to July 19. Each one was announced at the wire, not in advance. The support article contains no forward commitment beyond July 19 and no permanent inclusion decision. For teams deciding whether to build durable workflows on Fable 5, the practical takeaway is that plan-included access has now ended three different ways on paper in 18 days. Treat July 19 as the current end date, checked against the live support article, not as a final one. Budget the week after it as if credits-only pricing applies, and let a fourth extension be upside rather than a plan.

After July 19, unless Anthropic extends again, Fable 5 drops out of plan weekly limits entirely. Continued access is usage credits only at $10 per million input and $50 per million output, the highest published rate for a generally available Claude model. Cache writes run $12.50 (5m) and $20 (1h), cache hits $1. The official pricing page confirms these rates. Fable 5's Mythos-class safeguards still apply: high-risk cybersecurity and biology queries are automatically routed to Opus 4.8, not charged at Fable prices, and all Mythos-class traffic carries 30-day mandatory data retention for safety monitoring.

GPT-5.6 Sol gets inference optimizations, 10% more usage, and a temporary limit lift

OpenAI shipped a batch of GPT-5.6 Sol changes on July 13. Tibo Sottiaux from the OpenAI Codex team posted on X that the company has "landed inference optimizations and are passing down savings to all the subscriptions for GPT-5.6 Sol," which should result in around 10% more usage on its own. This is an effective capacity increase, not a per-token price cut. The API rate card is unchanged: Sol at $5/$30, Terra at $2.50/$15, Luna at $1/$6 per million tokens, all with a 1,050,000-token context window.

The GPT-5.6 launch generated roughly twice the previous peak traffic level, and OpenAI has been playing catch-up since. Over the weekend the company repeatedly reset usage limits for ChatGPT Work and Codex accounts. It also temporarily removed the five-hour usage-limit restriction for ChatGPT Plus, Business, and Pro subscribers, with no stated return date.

OpenAI is also expanding its banked reset feature, which lets users save a usage-limit reset and activate it when needed. The feature was initially desktop-only but is now accessible from the web and mobile apps. OpenAI added a banked reset to around 500,000 ChatGPT Work and Codex users and fixed a bug that prevented resets from applying correctly for a small percentage of users.

The fine print worth noting: these are subscription-side capacity adjustments, not API pricing changes. If you are building against the API, your per-token costs are the same. The 10% usage gain applies to ChatGPT, Codex, and Work subscribers, not to API callers. The temporary five-hour limit removal is exactly that, temporary, and OpenAI has not said when it ends.

Tracking: four deadlines still ahead

Gemini 3.5 Pro, July 17 (4 days, unconfirmed). Multiple third-party outlets report July 17 as the target GA date, but Google has not officially confirmed it with a model card, API documentation, or pricing. Leaked specs point to a 2-million-token context window and a Deep Think reasoning mode. Analyst estimates put API pricing near $15/$60 per million tokens, roughly 10x Gemini 3.5 Flash, but these are projections, not confirmed rates. The public Gemini API still lists only gemini-3.5-flash and gemini-3.1-pro-preview. Google has already slipped this model twice. Treat July 17 as a target, not a commitment.

DeepSeek V4 official launch, mid-July, and deepseek-chat retirement, July 24 (11 days). DeepSeek's official pricing page confirms the legacy aliases deepseek-chat and deepseek-reasoner retire on July 24, 2026 at 15:59 UTC. They currently route to deepseek-v4-flash; after the cutoff they return a hard error. The official V4 launch, which brings peak/off-peak surge pricing (2x during Beijing business hours), is still scheduled for mid-July with no exact date confirmed. Current off-peak rates: V4-Pro at $0.435/$0.87, V4-Flash at $0.14/$0.28 per million tokens.

OpenAI deprecation wave, July 23 (10 days). The official deprecations page lists gpt-5-chat-latest, gpt-5.1-chat-latest, five Codex variants, computer-use-preview, and deep-research models for retirement on July 23. GPT-5.4 base model is NOT on the list, despite a blog claim that circulated last week.

Sonnet 5 price jump, September 1 (7 weeks). Introductory pricing of $2/$10 runs through August 31. On September 1 it moves to $3/$15, a 50% input increase. The newer tokenizer produces approximately 30% more tokens for the same text, making the effective increase closer to $3.90/$19.50, which is more than Sonnet 4.6 at $3/$15 with the old tokenizer.

Current prices: what the ~11 most-compared models cost

Every rate below is crawled from the official pricing or model page, not an aggregator. All prices are per 1M input/output tokens in USD. Cached input listed where it materially changes cost.

OpenAI (developers.openai.com, crawled Jul 13):

• GPT-5.6 Sol: $5 / $30, 1.05M context, cached input $0.50
• GPT-5.6 Terra: $2.50 / $15, 1.05M context
• GPT-5.6 Luna: $1 / $6, 1.05M context

Anthropic (platform.claude.com, crawled Jul 13):

• Claude Fable 5: $10 / $50, cache hits $1, cache writes $12.50 (5m) / $20 (1h)
• Claude Opus 4.8: $5 / $25, cache hits $0.50
• Claude Sonnet 5: $2 / $10 through Aug 31, then $3 / $15 on Sep 1 (new tokenizer adds ~30% more tokens)

Google (ai.google.dev):

• Gemini 3.5 Flash: $1.50 / $9, 1.05M context. Gemini 3.5 Pro pricing not yet published.

xAI / SpaceXAI (docs.x.ai, crawled Jul 13):

• Grok 4.5: $2 / $6, 500K context. Grok 4.3: $1.25 / $2.50, 1M context. Batch discount 20% on 4.3/4.20 only, not 4.5.

Meta (dev.meta.ai, crawled Jul 13):

• Muse Spark 1.1: $1.25 / $4.25, cached input $0.15, 1M context. First paid Meta model. $20 free credits for new accounts. US-only waitlist.

DeepSeek (api-docs.deepseek.com, crawled Jul 13):

• DeepSeek V4-Pro: $0.435 / $0.87, cache hit $0.003625, 1M context. Peak-hour 2x surcharge coming at official launch.
• DeepSeek V4-Flash: $0.14 / $0.28, cache hit $0.0028, 1M context.

Mistral (mistral.ai):

• Mistral Large 3: $0.50 / $1.50, 256K context, open-weight (Apache 2.0).

Prices verified against official pages on July 13, 2026. Check the linked page before production budgeting, especially for models with upcoming changes (Sonnet 5 Sep 1, DeepSeek V4 peak pricing).