Mistral AI said on X on 28 October 2026 that it is “now serving” DeepSeek V4.1 Flash, in public preview, and Z.ai’s GLM 5.3 Flash, which it called “now generally available”. In the same thread, the French AI lab introduced a Responses API in public preview, “made for building agents on Mistral”.
Mistral’s price list puts DeepSeek V4.1 Flash at $0.30 per million input tokens and $1.20 per million output tokens, against $1.36 and $4.18 for its own Mistral Large 4. On Artificial Analysis’s Intelligence Index v5, DeepSeek V4.1 Flash 1017 (max) scores 32 and Mistral Large 4 Preview scores 25. By our arithmetic, the DeepSeek model costs 22% of Large 4’s input price and 29% of its output price on Mistral’s own list.
Short explainer · Mistral and DeepSeek brief
A vertical news Short narrated from the published headlines and opening lines of this piece and three related FourWeekMBA pieces on AI model scores, prices and model swapping.
Business Pill · Model routing
An evergreen Business Engineer explainer on model routing: sending each request to the cheapest model that can handle it. It explains the general idea only and does not discuss Mistral, DeepSeek or Z.ai.
The key insight: As we read it, Mistral is now selling an outside model that Artificial Analysis scores above Mistral’s own Large 4, at under a third of Large 4’s list price, through an API that uses OpenAI’s Responses format.
What happened next (30 Oct): Mistral released the Mistral Large 4 weights on Hugging Face on 29 October. The model card describes “a 1 trillion-parameter natively multimodal Mixture-of-Experts model with 49 billion active parameters per token” and says it is released under the Mistral Ecosystem Open-Weight License, under which “operating a Model-as-a-Service business requires a separate commercial license with Mistral.”
What Mistral Announced

Mistral wrote that each response from the new API “returns the model’s answer, its reasoning and its tool calls as separate items”, and that it “works with all chat models hosted by Mistral, including Large 4.” Its documentation says the API lets developers build “using the OpenAI Responses format” and that it “is stateless”, so an application keeps the conversation history and sends it with each request.
- DeepSeek V4.1 Flash 1017. Mistral’s model page, dated 28 October 2026, lists it as a public preview and a “third-party open weight text model from DeepSeek, hosted by Mistral”, under an MIT licence, with a 1M-token context window and 128k maximum output. The page says “The model is served without Mistral modifications.”
- Z.ai GLM 5.3 Flash. Mistral’s page, dated 22 October 2026, lists it as generally available and describes it as “a multimodal 320B-parameter (18B active) Mixture-of-Experts model”, also under an MIT licence and also “served without Mistral modifications”. Its price is $0.15 per million input tokens and $0.50 per million output tokens.
- Mistral’s posts on X. Mistral called DeepSeek V4.1 Flash “Great for long-context coding and agentic tasks” and GLM 5.3 Flash “Great for subagents, tool calls and coding tasks.”

Update (30 Oct): On OpenRouter’s public model API at 16:36 UTC on 30 October, Mistral Large 4 (listed as mistralai/mistral-large-4-0) had 3 endpoints, all served by Mistral. GLM-5.3 had 49 endpoints from many providers, Mistral among them (our count of the API listing).
What Was Posted on X
Mistral AI posted that it is introducing the Responses API in public preview, made for building agents on Mistral, returning the model’s answer, reasoning and tool calls as separate items, and that it works with all chat models hosted by Mistral, including Large 4.
In a follow-up post, Mistral AI said it is also now serving two fast, low-cost models: DeepSeek V4.1 Flash in public preview, and GLM 5.3 Flash, now generally available.
Adam Holter writes that Mistral hosts DeepSeek and GLM on OpenRouter, and that nobody else hosts Mistral’s open models.
Mistral Already Hosted Z.ai
The third-party catalogue predates this week. Mistral’s changelog says Z.ai GLM 5.3 “is now Generally Available” under a 28 September entry, and that the older GLM 5.2 retires on 31 October 2026. Its 22 October entry says GLM 5.3 Flash “is now available”. The changelog we read lists no DeepSeek model, and the DeepSeek model page is dated 28 October.
How the Prices Compare
Mistral’s API pricing page lists these prices per million tokens, input then output:
- Z.ai GLM 5.3 Flash: $0.15 and $0.50
- DeepSeek V4.1 Flash 1017: $0.30 and $1.20
- Mistral Small 4: $0.15 and $0.60
- Mistral Large 4: $1.36 and $4.18
- Z.ai GLM 5.3: $1.40 and $4.40
- Mistral Medium 3.5: $1.50 and $7.50
DeepSeek’s own API page lists the same model at $0.30 per million input tokens and $1.20 per million output tokens at peak, and says “Off-peak rates are half of the peak rates.” It sets peak hours at 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday. By our reading, Mistral’s price matches DeepSeek’s peak rate and is twice its off-peak rate. DeepSeek’s page also lists a 384K maximum output, against 128k on Mistral’s page.
Mistral separately offers regional endpoints for the EU and the US at “1.1× standard list pricing”, and its documentation says “Available models vary by region.”
What the Benchmark Shows
On Artificial Analysis’s leaderboard, read on 28 October, Claude Opus 5.5 (max with fallback) leads the Intelligence Index v5 at 58. Claude Haiku 5.5 (max) scores 36, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) each score 32, and Mistral Large 4 Preview scores 25. The leaderboard shows no Intelligence Index score for GLM-5.3-Flash.
The comparison has limits. Mistral’s page describes Large 4 as “a state-of-the-art, open-weight, general-purpose multimodal model” with “52B active parameters and 1.05T total parameters”, while it describes the DeepSeek model as a text model. The scores are for the settings Artificial Analysis lists, and both models are in preview on Mistral.
Connecting the Coverage
Artificial Analysis launched version 5 of its index on 28 October, with Claude Opus 5.5 (max) leading at 58, as we reported that morning. What’s new: Mistral now sells one of the open-weight models that ranks above its own flagship on that index.
DeepSeek wrote on 20 October that a single cluster behind DeepSeek V4.1 Flash serves 155 trillion tokens a day at US$38 per GPU-hour, as we covered on 25 October. Microsoft’s Satya Nadella wrote on 21 October that in Microsoft’s system “models become interchangeable”, with a catalogue of more than 11,000 models including Mistral’s, as we reported on 21 October. OpenAI priced GPT-6.1 Terra at $1 and $5 per million tokens on 26 October, as we covered.
The Structural Read
As we read it, the Responses API is built to make switching easier: Mistral’s documentation says it works whether you are starting a new application or “bringing an existing Responses integration to Mistral”.
Mistral’s price for DeepSeek V4.1 Flash matches DeepSeek’s own peak rate, and DeepSeek charges half that outside peak hours, so by our reading a buyer going through Mistral pays DeepSeek’s peak rate at all hours.
Mistral’s changelog shows it was hosting Z.ai models before this week, with GLM 5.3 generally available since 28 September; the DeepSeek model page is dated 28 October.
Mistral AI on X, 28 October 2026
“Each response returns the model’s answer, its reasoning and its tool calls as separate items, so your app can handle every step more easily than with Chat Completions.”
Three Implications
FOR DEVELOPERS Mistral’s documentation says the Responses API uses the OpenAI Responses format and is stateless, so the application keeps the conversation history and sends it with each request.
FOR DEEPSEEK Its V4.1 Flash model is listed on Mistral’s API at the same price as DeepSeek’s own peak rate, served “without Mistral modifications”, per Mistral’s model page.
FOR BUYERS COMPARING PRICES On Mistral’s list, the DeepSeek model costs $0.30 and $1.20 per million tokens against $1.36 and $4.18 for Large 4; Artificial Analysis scores them 32 and 25.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Open vs Closed Meta-Framework.
The framework puts it as “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read it, Mistral listing rival open-weight models next to its own treats models as one of the layers it resells through its API. The framework was published in March 2026 and does not discuss Mistral, DeepSeek or Z.ai.
What Is Not Established
The details here come from Mistral’s posts on X and its documentation. The pages we read do not say whether DeepSeek V4.1 Flash is available on the EU or US regional endpoints, whether Mistral has any agreement with DeepSeek or Z.ai beyond hosting their MIT-licensed weights, or how much traffic the third-party models carry.
Artificial Analysis’s score is for one model setting and may differ from results on Mistral’s hosted version. We did not contact Mistral, DeepSeek or Z.ai.
The Bottom Line
Mistral now serves DeepSeek V4.1 Flash at $0.30 and $1.20 per million tokens, against $1.36 and $4.18 for its own Mistral Large 4, and Artificial Analysis scores the DeepSeek model at 32 against 25 for Large 4 Preview. Mistral also opened a Responses API in the OpenAI format for building agents.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The announcement comes from Mistral AI’s posts on X on 28 October 2026; model details and prices come from Mistral’s documentation and DeepSeek’s API pricing page, read the same day. Benchmark scores come from Artificial Analysis’s leaderboard, read on 28 October. Percentages and price comparisons are our arithmetic. We did not contact Mistral, DeepSeek or Z.ai.
Sources: Mistral AI on X (Responses API), 28 Oct 2026 · Mistral AI on X (new models), 28 Oct 2026 · Mistral: DeepSeek V4.1 Flash 1017 model page · Mistral: Z.ai GLM 5.3 Flash model page · Mistral API pricing · Mistral changelog · Mistral Responses API docs · Mistral regional inference docs · DeepSeek API pricing · Artificial Analysis leaderboard, read 28 Oct 2026









