Mistral Medium 3.5 deprecated: what to use instead

Mistral Medium 3.5 deprecated: switch to Large 4

Updated 2026-10-31. Official Mistral docs, LMArena ranks.

Quick answer

Mistral deprecated Medium 3.5 on October 31, 2026 and tells new integrations to use Mistral Large 4 instead. Deprecated models keep working until their retirement date, then requests fail with a 404, so switch the model ID from mistral-medium-3-5 to mistral-large-4. The upgrade is also a price cut: Large 4 costs $0.70 input and $2.10 output per 1M tokens, against $1.50 and $7.50 for Medium 3.5, and it has a 1M context window instead of 256K.

	Mistral Medium 3.5	Mistral Large 4
Input price (per 1M)	$1.50	$0.70
Cached input (per 1M)	$0.15	$0.07
Output price (per 1M)	$7.50	$2.10
Context window	256K	1M
Status	Deprecated Oct 31, 2026	Public Preview (since Oct 6, 2026)
Model ID	`mistral-medium-3-5`	`mistral-large-4`
Open weights	Yes, Modified MIT	Yes, Mistral open weight license
Coding rank (LMArena)	#48 of 50	#29 of 50
Chat rank (LMArena)	#44 of 51	#43 of 51
Reasoning rank (LMArena)	#43 of 51	#26 of 51
Agents rank (LMArena)	#37 of 37	#32 of 37
Vision rank (LMArena)	#33 of 33	#21 of 33
Writing rank (LMArena)	#42 of 51	#45 of 51

USD per 1M tokens, Mistral API list prices. Large 4 checked 2026-10-31 on Mistral's model page; Medium 3.5 is no longer on Mistral's pricing page, so its row shows the last listed price. Ranks are from LMArena among the models we price.

What Mistral announced

Mistral's Medium 3.5 page now carries a "Deprecated" badge with the note "This model is deprecated. Use Mistral Large 4 for new integrations," a deprecation date of 10/31/2026 and Mistral Large 4 as the replacement. Medium 3.5 launched on April 28, 2026, so it lasted about six months as Mistral's mid-size API model.

Deprecated is not the same as switched off. Under Mistral's model lifecycle policy, a deprecated model "remains accessible" until its retirement date, and after that, requests to its identifiers "fail with a 404 error." Mistral gives 6 months' notice for General Availability models and 1 month for Public Preview models. The Medium 3.5 page doesn't show a retirement date yet, so treat today as the start of the clock, not the end.

Is Mistral Large 4 cheaper than Medium 3.5?

Yes, on every line. Input drops 53%, from $1.50 to $0.70; output drops 72%, from $7.50 to $2.10; cached input goes from $0.15 to $0.07. That's unusual: the replacement is the bigger model and the cheaper one. On our blended rate (3 parts input to 1 part output) Large 4 is $1.05 per 1M against $3.

Large 4 also ranks higher on LMArena on coding (#29 of 50 vs #48 of 50), reasoning, agents and vision. Writing is the one board where Medium 3.5 sits slightly ahead. The full side by side is on Mistral Large 4 vs Mistral Medium 3.5.

How to switch

Change the model ID. Replace mistral-medium-3-5 with mistral-large-4 in your API calls. Both serve the same chat completions, function calling, structured outputs, batching and agents endpoints, per their model pages.
Check aliases. If you call an alias like mistral-medium-latest, Mistral warns that aliases "may expose you to silent updates in model behavior and pricing." Pin a fixed ID so you know which model and price you get.
Know what Public Preview means. Large 4 is in Public Preview, which Mistral prices the same as General Availability but allows silent updates and only 1 month of deprecation notice. If you need a frozen model, test on a sample of real traffic and re-test after updates.
Re-run your evals. Large 4 is a different architecture (a mixture-of-experts model with 52B active and 1.05T total parameters, per Mistral). Output length and style can change, which moves your output-token bill too.
Self-hosting is a separate question. Medium 3.5 weights were released under a Modified MIT license, and the deprecation is about Mistral's API. Large 4 is open weight too, but at 1.05T parameters it's a server-scale model, not a laptop one.

What a month costs

List price, no discounts, our math. Medium is 20M input and 5M output tokens a month; heavy is 200M and 50M.

Model	Medium month	Heavy month
Mistral Medium 3.5	$67.50	$675
Mistral Large 4	$24.50	$245

Other Mistral and similar models

Mistral Large 4: every Mistral Large 4 price, cached rate and context detail.
Mistral Small 4: the cheaper Mistral option at $0.15 / $0.60 if Medium 3.5 was more model than you needed.
Mistral Large 4 vs DeepSeek V4 Pro: the open-weight flagship matchup.
All Mistral API prices and Mistral Le Chat limits explained.

Sources

FAQ

Is Mistral Medium 3.5 being shut down?

Mistral deprecated it on October 31, 2026. A deprecated model still works until its retirement date, then its IDs return a 404 error. Mistral hasn't posted the retirement date on the model page yet.

What replaces Mistral Medium 3.5?

Mistral Large 4. Mistral's Medium 3.5 page says to use Mistral Large 4 for new integrations. Its model ID is mistral-large-4.

Is Mistral Large 4 more expensive than Medium 3.5?

No, it's cheaper. Large 4 costs $0.70 input and $2.10 output per 1M tokens; Medium 3.5 was $1.50 and $7.50.

How long do I have to migrate?

Mistral's lifecycle policy gives 6 months' notice for General Availability models and 1 month for Public Preview models before retirement. Check the Medium 3.5 model page for the exact retirement date once Mistral posts it.

Is Mistral Large 4 production-ready?

It's in Public Preview. Mistral prices Preview models like General Availability ones, but Preview models can get silent updates and may be retired with 1 month's notice.

What context window does Mistral Large 4 have?

1M tokens, up from 256K on Medium 3.5, per Mistral's model page.