The eighth DeAI Price Index observation is dated 2026-11-09 UTC, one week after issue seven. Comparing OpenRouter's public catalog against the 2026-11-02 snapshot, eight of seventeen tracked rows moved — and for the second consecutive Monday they moved in mixed directions with no shared pattern. The headline: meta-llama/llama-4-maverick jumped 87% on input to $0.35/$1.00 after holding flat for three straight observations, the week's only mover against a "cheap drift" narrative. The catalog itself grew to 461 models from 452, and a new DeepSeek variant row appeared.
Key facts
- Llama 4 Maverick: $0.1875/$0.6525 → $0.35/$1.00 (input +87%, output +53%) — after three consecutive flat observations, the row jumped to its highest level in this index. The issue-seven story had named it part of the basket's "slowly stabilizing core." Observed on the OpenRouter models API.
- DeepSeek V4 Flash 0731: $0.119/$0.238 → $0.0434/$0.30 (input -63%, output +26%) — a fourth distinct shape in four observations ($0.44/$1.32 → $0.119/$0.238 → now $0.0434/$0.30), and the cheapest input leg the row has ever carried. It still sits below DeepSeek's own official peak card of $0.30/$1.20 (re-verified 2026-10-05).
- New catalog row: DeepSeek V4.1 Flash 1017 at $0.006/$0.50 (batch $0.011/$0.20) — the first appearance of the 1017 checkpoint as its own catalog entry, alongside the parent
deepseek-v4.1-flashrow, which itself moved to $0.051/$0.567 (+13% on both legs). Four observations of this family have now produced four different shapes. - DeepSeek V4 Pro: $1.60/$3.20 → $0.95526/$1.91052 (both -40%) — last week's +67% spike fully reverted to the exact row the index carried through issues 1–5. The spike lasted one observation.
- GLM-5.3 Flash: $0.0285/$0.0923 → $0.0335/$0.083 — the record-low row from issue seven scissored slightly (input +17%, output -10%) rather than holding; its two-observation test failed.
- Kimi K2.5 output +33% to $3.325; Kimi K3 output +7% to $16.00; GLM-5.3 output -61% to $1.40. Nine rows held flat, including both closed anchors (claude-sonnet-4.6 $3.00/$15.00, gpt-5.5 $5.00/$30.00). Hermes 4 70B is absent an eighth consecutive observation.
What happened
For the second Monday running, the diff has no single story line: movers went both directions, and the week's largest move hit a row that had looked settled. llama-4-maverick — flat at $0.1875/$0.6525 through issues 5, 6, and 7, and cited last week among three rows that "had not moved in at least three observations" — jumped to $0.35/$1.00 with no first-party Meta announcement. Meanwhile deepseek-v4-pro fully unwound its own one-week-old +67% spike, landing back on the exact $0.95526/$1.91052 row the index recorded from issue 1 through issue 5.
The two-observation discipline caught two more casualties. GLM-5.3-flash's issue-seven record low ($0.0285/$0.0923) did not hold a second Monday — it scissored to $0.0335/$0.083. And deepseek-v4-flash-0731's issue-seven low broke in the other direction: input fell another 63% to $0.0434 while output rose 26% to $0.30, a fourth shape in four weeks that matches no published DeepSeek card. The one row that did pass a stability test is trivially the most stable: deepseek-v4-flash (0423) held $0.091/$0.182 for a third consecutive observation.
The catalog grew for the first time in three weeks: 452 → 461 models. The growth is partly DeepSeek restructuring — the new deepseek-v4.1-flash-1017 entry plus a :batch variant and a ~deepseek/deepseek-v4-flash-latest alias row — not broad re-listing. A full catalog diff remains follow-up work beyond this story's basket.
Why it matters
- "Stable" rows break too, and they break upward. The strongest new datum this week is directional: the basket's longest-held stable row moved, and it moved +87%, not down. Weeks of open-model price coverage have emphasized the floor drifting lower; this week's largest mover ran the other way. A builder who anchored Maverick workloads to the $0.1875 row now pays 1.87× per input token on the same aggregator listing.
- The DeepSeek V4.1 Flash family still has no settled price — but it now has a structure. After four distinct parent-row shapes in four observations, the catalog finally shows a checkpoint-split family: parent at $0.051/$0.567, the new 1017 entry at $0.006/$0.50, batch legs around $0.011/$0.20. Whether the 1017 row holds two Mondays is the first test of whether this family's churn is converging or just rebranding. Until a first-party page confirms any of it, the rows stay unverified for tracker purposes.
- One-week spikes are noise, and the index now has receipts.
deepseek-v4-prospiked +67% in issue seven and fully reverted in issue eight;glm-5.3-flashhit a record low in issue seven and scissored away in issue eight. Across eight observations, no single-observation move on a volatile row has yet predicted the next Monday.
Background
The DeAI Price Index is a weekly, dated, append-only observation of list pricing — it records what a catalog listed on a date, and never rewrites prior issues. Issue one established the basket and methodology: OpenRouter's GET /api/v1/models prompt/completion fields converted to dollars per million tokens, batch rows excluded, plus first-party anchors. Issue seven scattered issue six's synchronized reversion; this issue mixes directions again, confirming that scattered movement — not reversion, not convergence — is this catalog's current norm.
A process note, per the standing gap report: the automated DIPT pricing fetcher again produced no machine-readable OpenRouter candidates in this week's sweep (data/tracker/resolved_2026-11-09.json carries a NEAR AI self-reported pricing post and a Lium GPU-rental claim, neither with a machine-readable pricing source configured), so the snapshot was fetched directly for this job using the documented Issue-1 methodology — the seventh week running without a wired OpenRouter adapter. The raw pull is dated and archived (data/tracker/raw_openrouter_models_2026-11-09.json, 461 models fetched live 2026-11-09 ~13:00 EET; basket extraction in pricing_fetch_2026-11-09.json); nothing here derives from provider marketing claims.
A second data note: the machine-readable snapshot file (/data/prices.json) had not been updated for issue seven — its snapshot was still dated 2026-10-26 when this issue published. That gap is corrected in this update: the 2026-11-09 snapshot is live, and the 2026-11-02 and 2026-10-26 snapshots were restored into the file's archive chain from that week's archived raw pulls, so the per-observation history is complete back to issue two.
What's next
The next observation is scheduled for Monday, 2026-11-16. Watch items: whether llama-4-maverick's +87% holds a second Monday (its first test since breaking), whether the new deepseek-v4.1-flash-1017 row survives and holds $0.006/$0.50, whether deepseek-v4-flash-0731's $0.0434 input leg finds any floor, whether kimi-k2.5's output +33% sticks, and whether the catalog's growth continues or the DeepSeek restructuring consolidates. The machine-readable snapshot updates at /data/prices.json alongside this story, with the 2026-11-02 snapshot restored in the file's archive chain.
Questions
- What changed in LLM API prices this week?
- On OpenRouter's public catalog, comparing 2026-11-09 to 2026-11-02: eight of seventeen Price Index rows moved, in mixed directions. llama-4-maverick rose 87% on input to $0.35/$1.00 after three flat weeks, deepseek-v4-flash-0731 fell to $0.0434/$0.30, deepseek-v4-pro gave back its prior 67% spike, and kimi-k2.5 output rose 33% to $3.325.
- What is DeepSeek V4 Flash 0731's price on OpenRouter now?
- The catalog row fell to $0.0434 input / $0.30 output per million tokens on 2026-11-09, from $0.119/$0.238 a week earlier — the fourth distinct shape in four observations, and the cheapest input leg the row has carried. DeepSeek's official peak list rate remains $0.30/$1.20 (re-verified 2026-10-05); the row is the aggregator's, not an official list price.
- Did DeepSeek release a new V4.1 Flash variant?
- A new deepseek-v4.1-flash-1017 row appeared in the OpenRouter catalog at $0.006/$0.50 per million tokens, with a batch variant at $0.011/$0.20. The parent deepseek-v4.1-flash row also moved, to $0.051/$0.567. No first-party DeepSeek pricing page change has been verified for either row; both remain aggregator observations.
- Did Kimi K3's output price keep rising?
- Yes, slightly. Moonshot AI's Kimi K3 listed at $0.192 input / $16.00 output per million tokens on 2026-11-09, versus $0.192/$15.00 a week earlier (+6.7% on output). The row has oscillated between $15.00 and $20.00 on output since mid-October with no first-party announcement.
Sources
- OpenRouter API — list models — OpenRouter
- DeepSeek API Docs — Models & Pricing — DeepSeek
- Open-Model Inference Price Deltas (Issue 7, 2026-11-02) — DeAI News
- Open-Model Inference Price Deltas (Issue 6, 2026-10-26) — DeAI News
About DeAI
DeAI is an independent publication covering open-weight AI models, private inference, and decentralized infrastructure — the tools for running AI you actually control. We test providers on price, privacy, and refusal behavior and publish the numbers, not the vibes. DeAI is powered by Morpheus (mor.org), a decentralized inference marketplace, and covers it on the same terms as every other provider.
Powered by Morpheus and StrandCMS
Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider — we rank it wherever the data lands. StrandCMS is the open-source, agent-first framework this site is built on.
