Independent/Reader-funded/Infrastructure, not tokens
DeAINEWS

AI you control — open models, private inference, and the networks that run them.

Industry & Business

Price Index Delta: Six Rows Scissor, DeepSeek Reshapes Flash

OpenRouter list-price delta, 2026-11-16: 6 of 17 rows moved — deepseek-v4.1-flash input jumped 175% to $0.14 while kimi-k2.5 fell to $0.50/$2.50.

DeAI is powered by Morpheus (mor.org). We cover competing providers on the same terms — see our methodology.

A single analog dial thermometer mounted on a cool gray datacenter corridor wall, its needle settled mid-scale after a visible swing, with a row of dark server racks receding down the corridor — a week in which moved catalog rows scissored rather than climbed. Illustration: DeAI
A single analog dial thermometer mounted on a cool gray datacenter corridor wall, its needle settled mid-scale after a visible swing, with a row of dark server racks receding down the corridor — a week in which moved catalog rows scissored rather than climbed. Illustration: DeAI

The ninth DeAI Price Index observation is dated 2026-11-16 UTC, one week after issue eight. Comparing OpenRouter's public catalog against the 2026-11-09 snapshot, six of seventeen tracked rows moved — and for the first time in this index's run, every mover scissored: each moved row pushed its input and output legs in opposite directions rather than up or down together. The largest input jump is deepseek/deepseek-v4.1-flash at $0.14/$0.42 (+175% on input, −26% on output), while moonshotai/kimi-k2.5 gave back most of last week's spike. The headline stability story is the opposite one: llama-4-maverick's +87% break held exactly, passing its first two-observation test. The catalog grew to 468 models from 461.

Key facts

  • DeepSeek V4.1 Flash: $0.051/$0.567 → $0.14/$0.42 (input +175%, output −26%) — a fifth distinct shape in five observations for this family. The catalog now also lists the separate deepseek-v4.1-flash-1017 entry at $0.15/$0.60 (batch $0.037/$0.33), where it was $0.006/$0.50 a week ago. Observed on the OpenRouter models API.
  • DeepSeek V4 Flash 0731: $0.0434/$0.30 → $0.06/$0.18 (input +38%, output −40%) — a fifth distinct shape in five observations for this row too; last week's record-cheap $0.0434 input leg did not survive a second Monday.
  • Kimi K2.5: $0.532/$3.325 → $0.50/$2.50 (input −6%, output −25%) — last week's +33% output increase reverted after exactly one observation, matching the index's standing pattern that single-observation moves on this row do not predict the next Monday.
  • DeepSeek V4 Pro: $0.95526/$1.91052 → $0.90/$1.50 (input −6%, output −21%) — now below the $0.95526/$1.91052 row the index carried through issues 1–5.
  • GLM 5.3 Flash: $0.0335/$0.083 → $0.0399/$0.0784 (input +19%, output −6%) — the Z.ai row scissored around the $0.03–$0.04 band for a third straight week; a sibling glm-5.3-flashx entry appeared at the identical $0.0399/$0.0784.
  • Eleven rows held flat, including llama-4-maverick at its new $0.35/$1.00 level (its break now two observations old) and both closed anchors (claude-sonnet-4.6 $3.00/$15.00, gpt-5.5 $5.00/$30.00). Hermes 4 70B is absent a ninth consecutive observation.

What happened

The diff finally has a shape, and it is not a direction — it is a scissors. All six movers cut one leg while raising or holding the other: deepseek-v4.1-flash (+175% in / −26% out), deepseek-v4-flash-0731 (+38% in / −40% out), kimi-k2.5 (−6% in / −25% out), deepseek-v4-pro (−6% in / −21% out), glm-5.3-flash (+19% in / −6% out), and gemini-2.5-flash was the exception only in magnitude (flat at $0.30/$2.50 — in fact it held; the sixth mover was deepseek-v4-flash 0423, which shaved to $0.09/$0.18, −1% on both legs). Nine weeks into this index, no observation has yet shown six simultaneous scissors moves; previous weeks scattered their movers across up, down, and reversion.

The DeepSeek Flash family is consolidating upward on input while fragmenting on structure. The parent v4.1 row now carries its highest input price in this index ($0.14), and the 1017 checkpoint row — which issue eight flagged at an eyebrow-raising $0.006/$0.50 — repriced to $0.15/$0.60, above the parent on input for the first time. Neither DeepSeek's official pricing page (last re-verified 2026-10-05, $0.30/$1.20 peak for the flash tier) nor any first-party announcement has confirmed any of these rows.

The week's stability datum is as important as the moves: llama-4-maverick held $0.35/$1.00 for a second consecutive observation after its issue-eight +87% break. That makes it the first upward break in this basket to survive the two-observation test — the same test that killed glm-5.3-flash's record low, deepseek-v4-pro's +67% spike, and kimi-k2.5's output jump.

Why it matters

  • Scissors moves hide real cost changes from both directions. A builder reading only "output prices fell" this week would miss that deepseek-v4.1-flash's input leg nearly tripled. For output-heavy workloads the DeepSeek family genuinely got cheaper; for long-context, input-heavy workloads it got more expensive. Leg-level tracking — the reason this index records input and output separately — is what catches it.
  • The 1017 row's $0.006 experiment is over. Issue eight noted the deepseek-v4.1-flash-1017 entry at $0.006/$0.50 as the family's cheapest-ever listing; one week later it sits at $0.15/$0.60, a 25× input repricing. The aggregator-catalog churn around this family has now produced five distinct parent-row shapes in five observations with no first-party confirmation for any of them — these remain unverified for tracker purposes.
  • One-observation moves are now formally unreliable — and one exception exists. kimi-k2.5's +33% output lasted one Monday; deepseek-v4-flash-0731's record $0.0434 input lasted one Monday. The two-observation discipline correctly discounted both. Maverick's $0.35/$1.00 is the first row to pass, which means the +87% is now the strongest single price signal this index has recorded on an open-weight anchor row.

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 eight mixed directions for a second week; this issue's uniform scissors pattern is new, and it further weakens any single-axis "open models are getting cheaper" narrative — the legs are diverging within rows even as the basket's flat rows (eleven of seventeen) keep their levels.

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, and no automated pricing-fetch file had landed by 13:25 EET, so the snapshot was fetched directly for this job using the documented Issue-1 methodology — the eighth week running without a wired OpenRouter adapter. The raw pull is dated and archived (data/tracker/raw_openrouter_models_2026-11-16.json, 468 models fetched live 2026-11-16 13:16 EET; basket extraction in pricing_fetch_2026-11-16.json); nothing here derives from provider marketing claims.

The machine-readable snapshot file (/data/prices.json) is updated alongside this story: the 2026-11-09 snapshot rotates into the archive chain, and the new snapshot carries the 2026-11-16 rows above. The catalog grew 461 → 468 this week, with additions including qwen/qwen4-max, anthropic/claude-fable-5.2, two new gemini-4 variants, and moonshotai/kimi-k3.1-pro:batch; thirteen rows were removed, mostly legacy checkpoints. A full catalog diff remains follow-up work beyond this basket.

What's next

The next observation is scheduled for Monday, 2026-11-23. Watch items: whether the DeepSeek family's input repricing holds a second Monday (deepseek-v4.1-flash at $0.14 is the week's largest move), whether the 1017 row stabilizes at $0.15/$0.60, whether Maverick's confirmed level becomes a third-observation anchor, whether kimi-k2.5 keeps drifting toward its pre-spike $2.50 output, and whether the scissors pattern repeats or reverts to scattered movement. The machine-readable snapshot updates at /data/prices.json alongside this story.

Questions

What changed in LLM API prices this week?
On OpenRouter's public catalog, comparing 2026-11-16 to 2026-11-09: six of seventeen Price Index rows moved, and every mover scissored — input and output legs moving in opposite directions. deepseek-v4.1-flash jumped from $0.051/$0.567 to $0.14/$0.42 per million tokens, deepseek-v4-flash-0731 to $0.06/$0.18, kimi-k2.5 fell to $0.50/$2.50, and deepseek-v4-pro to $0.90/$1.50.
Did DeepSeek V4.1 Flash's price go up?
On OpenRouter the row's input leg rose from $0.051 to $0.14 per million tokens (+175%) while output fell from $0.567 to $0.42 (−26%). No first-party DeepSeek pricing-page change has been verified; the row remains the aggregator's observation, not an official list price. The catalog also now lists a deepseek-v4.1-flash-1017 entry at $0.15/$0.60.
Did Kimi K2.5 get cheaper on OpenRouter?
Yes. The row fell from $0.532/$3.325 to $0.50/$2.50 per million tokens on 2026-11-16 (output −25%) — giving back most of the +33% output increase recorded on 2026-11-09, which had lasted exactly one observation. Moonshot AI published no first-party price change.
Did Llama 4 Maverick's price increase stick?
Yes. After jumping 87% on input to $0.35/$1.00 on 2026-11-09 — its first move after three flat weeks — the row held that exact level for a second consecutive observation on 2026-11-16, the first one-week upward move in this index to pass the two-observation test since issue two.

Sources

  1. OpenRouter API — list models — OpenRouter
  2. DeepSeek API Docs — Models & Pricing — DeepSeek
  3. Open-Model Inference Price Deltas (Issue 8, 2026-11-09) — DeAI News
  4. Open-Model Inference Price Deltas (Issue 7, 2026-11-02) — 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.

Learn more about the Morpheus Inference API →

Sponsor disclosure — not editorial

Powered by Morpheus and StrandCMS. Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider. StrandCMS is the open-source, agent-first framework this site is built on.

Learn more →