Listed prices. A dash means no single price is listed, or, under capability, no independent score.

API prices by model
ModelInputCache readOutputCapability (Epoch AI)
DeepSeek-V4-Flash-0731RetiredStale priceNo single priceNo single priceNo single priceNo independent capability score

Rates, billing conditions and sources

DeepSeek-V4-Flash-0731

Model ID: deepseek-ai/DeepSeek-V4-Flash-0731 · Context window: 1M tokens

Retired Nov 9, 2026 · price last observed Nov 8, 2026 (retirement notice ↗)

All rates, conditions and sources

Conditions of every rate below:

  • Nebius Token Factory public model catalog API (tokenfactory.nebius.com/api/public/models_info, which Nebius's model catalog calls "the authoritative machine-readable source"): input_price_per_million_tokens / output_price_per_million_tokens / input_cached_price_per_million_tokens, per 1 million tokens.
  • Model "deepseek-ai/DeepSeek-V4-Flash-0731" (the Nebius Token Factory API model ID), listed as an active text2text model with one flavor, labelled "cheap", and external_provider false.
  • As listed: serving region us-central1; quantization fp8; weights repository https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731. Hosts may serve different builds of a model; this listing is not proof of the developer's own weights.
  • Currency: the API names none; Nebius's Token Factory documentation states model-card prices "in USD per million tokens".
  • List prices: the API states no endpoint, account eligibility or regional price difference. It lists a cached-input price for this model; when cached input applies is not stated. Dedicated endpoints, post-training and embeddings are out of scope.

Input: $0.14 per 1M tokens · Standard tier

  • Input tokens (input_price_per_million_tokens).

Output: $0.28 per 1M tokens · Standard tier

  • Output tokens (output_price_per_million_tokens).

Cache read: $0.01 per 1M tokens · Standard tier

  • Cached input tokens (input_cached_price_per_million_tokens, which Nebius's model catalog lists among its token-model pricing fields). Nebius states no cache eligibility, TTL or cache-write price; they are unknown.

Price observed: · Price freshness: stale

API availability: not verified · Retired: Nov 9, 2026

Collection record

Raw API evidence ↗ · provider API · Claims: nebius-deepseek-v4-flash-0731: specifications · Retrieved · Checked · Reviewed · Freshness: unknown · Confidence: B

  • Nebius Token Factory public model catalog API (models_info), model_description of the flavor "deepseek-ai/DeepSeek-V4-Flash-0731", raw SHA-256 279526ca433255dc4e19562886346dcbc8c8d34f8bdfaf577b1713ad28135b17: "DeepSeek V4 Flash 0731 is a 1M-context reasoning model designed for coding and agentic workloads."
  • Shorthand stated in Nebius's own model description, recorded as stated under #504 (owner decision 2026-10-24, #622); not converted to an exact token count. Description prose, not a context field, hence confidence B.
  • Specifications only: no pricing, availability, capability or release-date assertion.

Primary source ↗ · provider documentation · Claims: nebius-deepseek-v4-flash-0731: availability · Retrieved · Checked · Reviewed · Freshness: unknown · Confidence: A

  • Nebius Token Factory documentation, "Model Deprecations: November 2026", raw SHA-256 9fd3f945fe2b9a357c243ec57203dd2b025bbbcc501bcaccaaabf88a9f3563de (retrieved 2026-11-06T16:31:55Z): "On November 9, 2026: Affected Serverless models will no longer be available via API or Playground. Dedicated Endpoints for these models will be unaffected by this change." and "These models will be removed from Token Factory Serverless on November 9, 2026. After that date, you will not be able to select them in the Playground or send API requests using their model IDs."
  • The notice's "Deprecating Model" table lists deepseek-ai/DeepSeek-V4-Flash-0731 with recommended replacement deepseek-ai/DeepSeek-V4.1-Flash-1017. It states: "Token Factory does not automatically reroute requests from deprecated models."
  • The published prices are Serverless list prices from the models_info API, the usage the notice covers; Dedicated Endpoints are out of scope.
  • Availability stays available: the models_info API still listed the model as active when this was reviewed (2026-11-06), and the removal date is in the future, so no retirement is recorded before it happens. reviewAfter is the removal date, a re-review date, not a retirement.
  • Nebius's models_info API (raw SHA-256 c8226b0f1921d48f469005af4d5d228a6602dacabb3793ec9e28662a480913b5, 2026-11-06 daily archive) lists this flavor with "deprecated": false, and as active. The flag is undocumented and disagrees with the notice for this model (#962); the dated notice is the basis of this record.
  • The same notice text is in the documentation's llms-full.txt (raw SHA-256 99cee4c2b34501f641aeacc38cf8e8fba45b7a975af5e8f1f82123ad028c1cc2, retrieved 2026-11-06T14:31:08Z).
  • Reviewed by Claude (owner delegation 2026-09-27), #996.
  • Re-reviewed 2026-11-11 (#999), superseding the availability and reviewAfter statements above, which describe the 2026-11-06 review: Nebius removed deepseek-ai/DeepSeek-V4-Flash-0731 from the models_info API between the 2026-11-09 and 2026-11-10 collections, as the notice dated; the retirement is recorded by retirement:nebius-deepseek-v4-flash-0731:2026-11-11, so this record has no further review date. First reviewed 2026-11-06T16:33:57Z (#996).

Raw API evidence ↗ · provider API · Claims: nebius-deepseek-v4-flash-0731: pricing · nebius-deepseek-v4-flash-0731: availability · 56 more claims about 28 other catalog entries · Retrieved · Checked · Review due · Freshness: stale · Confidence: B

  • Raw SHA-256: 0f3c08ebef5a341541096adec8ac77252a6959aebf924f7fc0934f5feaecbae8; collector nebius-models-info-json-1.
  • Hosted listing and pricing observation only: the reviewed token models of the public catalog; no checkpoint, developer, license, context-window, modality or release-date assertions. Like the other hosted providers, a listing is not proof of identical weights or behavior: some listings name a quantized or third-party build.
  • The API names no currency; US dollars rest on Nebius's Token Factory documentation (model cards show prices "in USD per million tokens"), hence confidence B. Availability rests on Nebius's own API listing the model as active (with external_provider false).

Primary source ↗ · provider documentation · Claims: nebius-deepseek-v4-flash-0731: retirement · nebius-deepseek-v4-flash-0731: availability · Retrieved · Checked · Reviewed · Freshness: unknown · Confidence: A

  • Nebius Token Factory documentation, "Model Deprecations: November 2026", raw SHA-256 cbd50601dd7ab41cabffedef0a6f60753587e35db103b78339097a8007582ab1 (retrieved 2026-11-11T02:48:39Z): "On November 9, 2026: Affected Serverless models will no longer be available via API or Playground. Dedicated Endpoints for these models will be unaffected by this change." and "These models will be removed from Token Factory Serverless on November 9, 2026. After that date, you will not be able to select them in the Playground or send API requests using their model IDs." Its "Deprecating Model" table lists deepseek-ai/DeepSeek-V4-Flash-0731 with recommended replacement deepseek-ai/DeepSeek-V4.1-Flash-1017. The text is unchanged from the 2026-11-06 copy (raw SHA-256 9fd3f945fe2b9a357c243ec57203dd2b025bbbcc501bcaccaaabf88a9f3563de) that deprecation:nebius-deepseek-v4-flash-0731:2026-11-06 quotes.
  • Observed removal, in Nebius's models_info API (tokenfactory.nebius.com/api/public/models_info) as archived by the daily refresh: the 2026-11-09 collection (retrieved 2026-11-09T04:31:27.328Z, raw SHA-256 0f3c08ebef5a341541096adec8ac77252a6959aebf924f7fc0934f5feaecbae8, 44,234 bytes, 31 entries) listed deepseek-ai/DeepSeek-V4-Flash-0731 as active; the 2026-11-10 collection (retrieved 2026-11-10T04:29:59.019Z, raw SHA-256 75818db5c912d5249da37ada7349e0014cc66f996cd39b7c84bce7f746f1c157, 35,780 bytes, 25 entries) no longer lists it. The 6 entries that left are exactly the 6 notice models recorded as retired; none was added. The removal happened between those two collections.
  • retiredAt is the notice's date, November 9, 2026. The notice states no time or time zone, so it is stored as 2026-11-09T00:00:00Z and shown as the date only; the two collections above bracket the removal.
  • The rates kept are the last observed Serverless list prices (the 2026-11-08 record; the 2026-11-09 collection repeated its body byte for byte). Dedicated Endpoints, which the notice leaves unaffected, are out of scope. Availability is unknown, the delisting convention (#427/#433): no current listing remains to cite.
  • If Nebius lists deepseek-ai/DeepSeek-V4-Flash-0731 again, the collector fails for review (#999); a refresh never undoes or overwrites a retirement.
  • Reviewed by Claude (owner delegation 2026-09-27), #999.
About capability scores (Epoch AI)

Capability scores: Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks' [online resource], epoch.ai/benchmarks. Accessed Nov 18, 2026. Licensed CC BY 4.0; values here are rounded to one decimal place and grouped into bands.

The Epoch Capabilities Index (ECI) combines many benchmarks on a scale where Claude 3.5 Sonnet is 130 and GPT-5 is 150. Bands, on the displayed one-decimal score: Frontier 160+, Strong 150–159.9, Capable 140–149.9, Entry below 140. Each score has a 90% range, listed with each model's rates and sources; a range that crosses a band boundary is flagged there. Models are not ranked: neighbouring scores often overlap.

A benchmark is not your workload. Epoch's scores mostly reflect runs at the highest reasoning effort, which use more output tokens than default settings. Scores are dated and can shift between Epoch releases. A score belongs to the developer's model; a host listing shows the same model's score, but hosts may serve different versions or quantizations.

How headline rates are chosen

Headline rates use the directory's shared rule: the rate a buyer meets by default, chosen by billing, tier and context dimensions and never by price. Context caps, the bracket starting at zero and short-context variants are labeled. Every price is a listed price, not a service guarantee; each offer's details say what its source states about availability. Batch, off-peak and other variants remain in the details; where no single rate is the default, the table shows a dash.

No currency conversion is performed. Missing cache prices are unknown, not zero. Equal units do not establish equivalent models or service conditions. See the methodology and source register.