PriceIndex

Daily price report

AI Model Price Changes

On October 5 several tracked models moved materially: OpenAI’s GPT‑5.6 Sol Pro token rates show large per‑token cuts in our feed, DeepSeek published and documented a multi‑tier repricing tied to its V4 lineup, and trackers recorded sharp increases for some Meta and Z.ai routes. For many of the smaller changes (MoonshotAI, NVIDIA, Qwen, Google, Tencent) there is no direct vendor statement explaining the Oct 5 moves — we link the official rate pages where vendors published them and cite price‑tracker snapshots where they did not.

Price changes

27

Models repriced

13

Price cuts

14

Price increases

13

Major, public vendor updates

OpenAI — GPT‑5.6 Sol Pro: our price feed recorded large reductions to Sol Pro per‑token rates (example: input, output and cached‑input reductions visible in Oct 2026 snapshots). OpenAI has published pricing and product posts about GPT‑5.6 and earlier Sol promotions; however we did not find a single OpenAI blog post that explicitly states Oct 5 as the cause for the specific 50% Pro price moves recorded in the feed. See OpenAI’s GPT‑5.6 documentation for the vendor’s published pricing guidance. (openai.com)

DeepSeek — V4 family: DeepSeek published a new V4 rate card and release notes (DeepSeek‑V4.1 / V4‑Pro changes, peak/off‑peak tiers and a shift in Flash/Pro routing). DeepSeek’s public docs and blog posts explain the company’s move to a peak/off‑peak two‑tier rate card and the V4.1 Flash / V4‑Pro positioning, which account for many of the V4‑series price moves tracked around early October. (deepseek.ai)

Tracker‑reported moves with no vendor rationale published

MoonshotAI — Kimi K3: trackers show per‑route changes (including cached‑input and output reprice on Oct 5) but Moonshot’s public launch/pricing posts we found predate October and do not include an Oct 5 explanation. We could not find a MoonshotAI statement that explains the Oct 5 delta. (forum.moonshot.ai)

Meta — Llama 3.3 70B Instruct: price increases were recorded in multiple price trackers on Oct 5–6; Meta does not publish a public line‑item justification for that repricing in its public model pages, and we found only tracker snapshots documenting the change. (weaveprism.com)

NVIDIA — Nemotron 3.5 Lightning: small per‑token adjustments appear in tracker snapshots (input +1%, output −6%, tiny cached‑input change). NVIDIA’s product press material announces the Nemotron family but does not include a public Oct 5 pricing rationale for the small moves recorded by trackers. (nvidianews.nvidia.com)

Qwen — Qwen3.8 / Qwen3.6 (27B): OpenRouter and price trackers show small in/out adjustments in early October; we did not find a direct Qwen (Alibaba Qwen team) public changelog that explains the Oct 5 shifts. (writeworks.ai)

Google — Gemma 4 26B A4B: trackers recorded ~+33% host pricing moves in early Oct; Google publishes Gemma model pages and cloud/Vertex listings, but we did not find a Google statement specifically attributing the Oct 5 price deltas captured by trackers. (theknowngood.com)

Smaller adjustments and context

Tencent — Hy3 and Hy4 preview: Tencent’s published Hy3/Hy4 product pages and TokenHub region pages list vendor pricing for Hy3 and Hy4 preview. Trackers show the Hy3/Hy4 per‑token drops recorded on Oct 5, but we did not find a Tencent press release that calls out Oct 5 as the cause. Tencent’s model release pages and TokenHub docs remain the authoritative vendor references. (tencent.com)

Z.ai — GLM‑5.3: several trackers and aggregator pages show GLM‑5.3 route price increases around Oct 5. Z.ai’s published pricing (docs/pricing) documents the GLM family’s list card — we found tracker evidence of the Oct 5 move but no Z.ai blog post explicitly tying a product decision to the Oct 5 price deltas. (glm5.app)

Summary of the pattern we see: public vendor posts (DeepSeek, Tencent, OpenAI product pages) explain some of the larger structural changes (new rate cards, promotions, peak/off‑peak rules). For many per‑route changes captured on Oct 5 the only public evidence is real‑time price registries (OpenRouter/aggregators/price‑trackers) rather than a vendor press release; where a vendor has published an explanation we cite it, and where they have not we note that explicitly and point to the tracker snapshot.

Why providers changed prices

moonshotai

MoonshotAI (Kimi K3): price snapshots on Oct 5 show an increase to the model’s output token rate and a decrease in the cached‑input rate on some routes; Moonshot’s public launch/pricing pages list Kimi K3 pricing but we could not find a MoonshotAI announcement that explains the Oct 5 route‑level changes. For the vendor’s published baseline pricing and K3 launch materials see the Moonshot/Kimi K3 announcement and pricing pages. (forum.moonshot.ai)

openai

OpenAI (GPT‑5.6 Sol Pro): trackers show large per‑token reductions recorded in early October for Sol Pro (input/output/cached input). OpenAI has published product and pricing pages for GPT‑5.6 and prior promotional pricing, but we did not find a single OpenAI blog post explicitly saying the Oct 5 snapshot was caused by a new OpenAI policy change; OpenAI’s public GPT‑5.6 pages explain prior Sol pricing and promotional adjustments. (Official OpenAI model/pricing pages and changelog referenced.) (openai.com)

meta

Meta (Llama 3.3 70B Instruct): price trackers recorded substantial increases for many hosts on Oct 5–6. Meta publishes model cards and Llama documentation, but we did not find a Meta press release or pricing note attributing these Oct 5 tracker moves to a Meta decision. The changes are visible in multiple price‑tracking snapshots. (weaveprism.com)

nvidia

NVIDIA (Nemotron 3.5 Lightning): tracker snapshots show small input/output/cached‑input adjustments on Oct 5. NVIDIA’s Nemotron family launch materials describe the model family but we found no NVIDIA statement explicitly explaining the recorded Oct 5 micro‑adjustments. (nvidianews.nvidia.com)

tencent

Tencent (Hy3 / Hy4 preview): Tencent’s product pages and TokenHub region docs publish Hy3 and Hy4 preview list prices. Trackers recorded the Oct 5 per‑token reductions for Hy3/Hy4 routes in our dataset, but we did not find a Tencent press note directly saying 'this is why' for the Oct 5 feed moves. See Tencent’s Hy3/Hy4 release and TokenHub pricing pages for the vendor’s published rates. (tencent.com)

deepseek

DeepSeek (V4 series): DeepSeek published an updated V4 rate card, the DeepSeek‑V4.1 GA post and an explanation of peak/off‑peak tiers and the new Flash/Pro positioning — those vendor posts explain most of the V4 family’s repricing and the move to a two‑tier (peak/off‑peak) structure that affected the V4‑series rates tracked around Oct 5. (deepseek.ai)

z-ai

Z.ai (GLM‑5.3): GLM‑5.3 list prices are documented in Z.ai’s published pricing table; multiple trackers show host‑level price increases around Oct 5. We could not find a Z.ai blog post explicitly attributing the Oct 5 host‑level price moves to a published vendor decision. (glm5.app)

qwen

Qwen (Qwen3.8 27B / Qwen3.6 27B): OpenRouter and price trackers show small early‑October adjustments (tiny input upticks, output cuts on some routes). We did not find an Alibaba/Qwen press release explaining the Oct 5 feed changes; the only public evidence for the Oct 5 moves is in route snapshots and aggregator logs. (writeworks.ai)

google

Google (Gemma 4 26B A4B): host pricing snapshots on early October show higher per‑token host rates for Gemma 4 26B on some routes. Google’s Gemma model pages and cloud docs list vendor rates in their published model documentation, but we did not find a Google announcement explicitly explaining the Oct 5 host price changes captured by trackers. (tokencost.app)

Sources

Biggest movers

Largest cuts

Largest increases

All changes by provider

Meta1 model

DeepSeek3 models

Z.ai1 model

MoonshotAI1 model

Kimi K3Oct 5
Cached input$0.80$0.33-59%Output$13$14+8%

OpenAI1 model

GPT-5.6 Sol ProOct 5
Input$4.00$2.00-50%Output$20$10-50%Cached input$0.40$0.20-50%

Tencent2 models

Hy3Oct 5
Input$0.13$0.08-38%Output$0.53$0.33-38%Cached input$0.03$0.02-38%
Hy4 previewOct 5
Cached input$0.04$0.04-10%Input$0.83$0.75-10%Output$2.50$2.25-10%

Google1 model

Gemma 4 26B A4BOct 5
Cached input$0.04$0.05+33%Output$0.23$0.30+33%Input$0.07$0.09+33%

Qwen2 models

NVIDIA1 model

FAQ

Which AI models cut prices on October 5, 2026?

7 models cut at least one token price on October 5, 2026, 14 price cuts in total. The biggest drop was Kimi K3 cached input at -59%.

Did any AI model prices increase on October 5, 2026?

Yes — 13 prices across 9 models increased. The largest was Llama 3.3 70B Instruct input at +120%.

Did OpenAI publish a reason for the Oct 5 GPT‑5.6 Sol Pro price change recorded in the feed?

OpenAI has published pricing updates and product posts for GPT‑5.6 (including a prior Sol promotional reduction), but we did not find a single OpenAI blog post or changelog entry that explicitly states Oct 5 as the cause for the specific 50% Pro price moves recorded in the tracker snapshots. See OpenAI’s GPT‑5.6 pages for the vendor’s published guidance. (openai.com)

If a model’s route price changed on Oct 5 but the vendor didn’t publish an explanation, what should I do?

Treat the tracker snapshot as an observed route price (it reflects what some providers or gateways are charging). For budget‑sensitive workloads, check the vendor’s official pricing page and your provider dashboard or billing export (they are the contractual source). If you need a formal explanation, contact the vendor’s support or your provider account rep and ask for an explanation or invoice reconciliation.

Which vendors published official explanations for the Oct 5 moves?

DeepSeek published an updated V4 rate card and release notes explaining its new peak/off‑peak pricing and V4‑series positioning. Tencent, OpenAI, Z.ai and others have public model pages and prior pricing posts, but for many of the Oct 5 route‑level deltas we only found tracker snapshots rather than a dated vendor press release explicitly naming Oct 5 as the reason. (deepseek.ai)