The Biggest Vendor Pricing Moves This Week (November 5, 2026): Five Changes That Move Real Budgets
Five pricing changes landed in one week across OpenAI, Anthropic, Tencent, GitHub and Zhipu, and only one of them is a discount. Priced from each vendor's own page on November 5, 2026: OpenAI launched GPT-6 Astra at $10/$50 per million tokens with Fast and Ultrafast accelerator tiers and Batch at half price; Anthropic moved all four mainstream tiers onto the 5.x generation with cache hit multipliers that differ by tier (0.025x for Fable 5.1 and Mythos 5.1, 0.05x for Opus 5.5 and Sonnet 5.5, 0.1x for the rest); Tencent shipped Hunyuan Hy4 GA at CNY 6/18 with vision and computer operation added at no price increase; GitHub Copilot filled in its Pro+ and Max personal tiers at $39 and $100; and Zhipu withdrew the promotional tier on GLM-5.3-Flash, which returns to its standard CNY 0.80/2.80. Also covers the Double 11 subscription discounts at Alibaba Cloud and Baidu Qianfan, the off-peak credit multipliers at Volcengine, and a budget-impact table showing what each change does to a 100M-token monthly bill.
Price changes arrive in two shapes, and this week delivered one of each. The first is a new price for something you already buy: GPT-6 Astra slots in above the current OpenAI line, Claude 5.5 replaces what Anthropic was selling last month, and Copilot Pro+ and Max fill gaps that left heavy users with nowhere sensible to go. The second is a price that came back up: Zhipu's GLM-5.3-Flash promotion ended, and a lot of budgets were still built on the discounted number.
OpenAI's new entry is **$10 input / $50 output per million tokens**, cache at $1, 1.05M context and a 128K maximum output. The same page also lists **Fast at $20/$100** and **Ultrafast at $60/$300**, with **Batch at $5/$25**. Four models now sit in active sale - GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna - while GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini shut down on January 14, 2027. Astra is not a general upgrade; it is a different price band, and most workloads should stay on Luna at $0.10/$0.50.
Anthropic's four mainstream tiers all moved to the 5.x generation and all carry 1M context: Haiku 5.5 at $0.10/$0.50, Sonnet 5.5 at $2/$10, Opus 5.5 at $4/$20 and Fable 5.1 at $10/$50. The tier jump matters - Haiku 5.5 steps to $0.50/$2.50 past 100K input tokens - but the detail most likely to be misread is the cache hit multiplier, which now varies by tier. Fable 5.1 and Mythos 5.1 cache at 0.025x base input price, Opus 5.5 and Sonnet 5.5 at 0.05x, and every other model at 0.1x. For an agent that rereads the same context thousands of times, that multiplier moves the bill more than the headline input price.
Hunyuan Hy4 went GA on November 4 at **CNY 6 input / CNY 18 output per million tokens** - identical to the August preview, so this is capacity added rather than a price rise. The new part is capability: 770B total parameters with 49B active, 1M context, and **vision plus computer operation for the first time**, with dedicated training for Blender and CAD. For teams holding back on a multimodal workflow because domestic flagship prices were uncomfortable, Hy4 changes the arithmetic rather than the answer.
Zhipu's official pricing table now shows only the standard rate for GLM-5.3-Flash: **CNY 0.80 input / CNY 2.80 output / CNY 0.23 cache hit**, with no discount row at all. The earlier promotional tier is withdrawn. This is the change most likely to have already hit a budget: many cost comparisons still use CNY 0.40 / CNY 1.40 as the GLM-5.3-Flash figure. Batch API is unaffected - the rule that supported Batch models bill at 50% of standard list price still stands, so a Batch call now bills at exactly the old effective rate.
The genuine discounts this week sit on subscriptions and off-peak windows, not on list API prices. Alibaba Cloud Bailian Token Plan halves its entry tiers for new users during the Double 11 window - **Lite at CNY 19.50 against a CNY 39 list price, Essential at CNY 39.50 against CNY 79** - and adds nightly Credits discounts from 22:00 to 08:00 at 60% of standard for qwen3.8-max and qwen3.8-flash, 70% for the DeepSeek V4 family, and free for decision-model-preview. Baidu Qianfan stacks three offers: first purchase from CNY 4.9, off-peak calls at 20% between 22:00 and 08:00, and 40% off the first renewal.
Frequently Asked Questions
What is the single biggest pricing change of this week?
It depends on which side of the ledger you sit on. For a team that planned around GLM-5.3-Flash at CNY 0.40 per million tokens, the withdrawal of that promotional tier is the biggest change, because it doubles the input rate back to CNY 0.80 and CNY 1.40 per million output becomes CNY 2.80. For a team choosing where to put new workloads, GPT-6 Astra at $10/$50 and Claude Fable 5.1 at $10/$50 are the biggest change, because they establish a new top price band that did not exist a week ago. For a subscriber, the Alibaba Cloud Token Plan entry tiers at CNY 19.50 and CNY 39.50 for new users are the biggest change - that is a 50% cut on the cheapest domestic plan, and it expires on November 30.
Does the GLM-5.3-Flash price increase affect Batch API calls?
No. Zhipu's pricing page states that models supporting Batch API are billed at 50% of the corresponding standard call price, and that is a standing rule rather than a promotion. What ended was the discount on the standard interactive rate. So a Batch call on GLM-5.3-Flash now bills at 50% of CNY 0.80 input and CNY 2.80 output, which is CNY 0.40 and CNY 1.40 - numerically the same as the old promotional rate, but derived from the standard price rather than from a temporary offer. Interactive calls at full price cost double the old promotional input rate.
How much does the Claude cache multiplier difference save in practice?
The gap is a factor of four between the cheapest and most expensive cache tier. On Claude Fable 5.1 and Mythos 5.1, cache hit costs 0.025x base input price, so the $0.25 cache rate on Fable 5.1 comes from a $10 base input price. On Opus 5.5 and Sonnet 5.5 it is 0.05x, giving $0.20 on a $4 base and $0.10 on a $2 base. Every other model sits at 0.1x. Concretely, for a 100M-token workload with 80% cache hit rate, the cache component costs about $200 on Fable 5.1 versus about $800 on a 0.1x model - a $600 monthly difference for a workload that produces identical output. On Sonnet 5.5 the cache component would be roughly $80.
Should I switch providers now, or wait for the next price cycle?
Do not switch providers on the strength of one weekly change, because switching carries real costs that the price table does not show: prompt cache cold starts, re-tuning against a different tokenizer, and re-testing tool calling behavior. What the numbers do support right now is narrower. If you run GLM-5.3-Flash interactively and were budgeting on the promotional rate, adjust the budget immediately - that is a real cost change on the same model, not a migration. If you run non-interactive batch work, the Batch path already preserves the old effective price. And if you are picking a subscription, the Alibaba Cloud and Baidu Qianfan offers expire on November 30 and November 31 respectively, so those decisions have a real deadline while provider migrations do not.
Want more AI pricing comparisons?
Check our price changelog for daily updates on AI model pricing, new releases, and deprecations.