Benchmark report
Is the Mistral API slow?
Mistral Medium time to first token from 5 cities, measured daily
Uneven around the world
First visible token arrived in 245 ms from Amsterdam and 442 ms from Tokyo, with every city under half a second. The gap grows with distance from Europe, which suggests the model is served from there.
- Fastest
- 245 ms
- Amsterdam
- Slowest
- 442 ms
- Tokyo
- Successful requests
- 100%
- 5 of 6 locations
Response time around the world
Amsterdam245 ms
Montreal310 ms
San Francisco347 ms
Singapore373 ms
Tokyo442 ms
MumbaiCollecting since 8 Sept 2026
No thinking phase, so the stream opens with the answer
In the controlled run of 13 September the stream opened at the same moment the first visible token arrived, within a millisecond, in every city, and the one-word reply was complete 21 to 31 ms later. Connecting took 2 to 8 ms and the secure handshake 5 to 15 ms; the rest was waiting: 239 ms in Amsterdam, 296 in Montreal, 357 in Singapore, 364 in San Francisco and 435 in Tokyo. Every city reached the same Cloudflare address through a nearby point of presence (London for Amsterdam), so the wait is the trip from that edge to Mistral plus the time the model takes to produce its first token. About a quarter of a second of it is there even in Amsterdam, and the rest grows with distance from Europe.
Click the location to see each stage.
Tokyo468 ms
Tokyo- Finding the server
- 1 ms
- Reaching the server
- 2 ms
- Setting up security
- 9 ms
- Waiting for the server: most of the time
- 435 ms
- Receiving the response
- 22 ms
- Total
- 468 ms
- Response opened
- 446 ms
- First visible word
- 447 ms
- Answer complete
- 468 ms
Finding the server is measured once per location, so it sits outside these bars. Open a city to see it.
Bars show the typical request, so the parts add up to its total. Why
Compare all locationsHide the comparison
Click a city to see its stage-by-stage breakdown.
Amsterdam288 ms
Amsterdam- Finding the server
- 1 ms
- Reaching the server
- 8 ms
- Setting up security
- 15 ms
- Waiting for the server: most of the time
- 239 ms
- Receiving the response
- 26 ms
- Total
- 288 ms
- Response opened
- 262 ms
- First visible word
- 262 ms
- Answer complete
- 288 ms
San Francisco400 ms
San Francisco- Finding the server
- 10 ms
- Reaching the server
- 4 ms
- Setting up security
- 8 ms
- Waiting for the server: most of the time
- 364 ms
- Receiving the response
- 24 ms
- Total
- 400 ms
- Response opened
- 376 ms
- First visible word
- 376 ms
- Answer complete
- 400 ms
Montreal336 ms
Montreal- Finding the server
- 1 ms
- Reaching the server
- 3 ms
- Setting up security
- 5 ms
- Waiting for the server: most of the time
- 296 ms
- Receiving the response
- 32 ms
- Total
- 336 ms
- Response opened
- 304 ms
- First visible word
- 305 ms
- Answer complete
- 336 ms
Singapore395 ms
Singapore- Finding the server
- 2 ms
- Reaching the server
- 4 ms
- Setting up security
- 7 ms
- Waiting for the server: most of the time
- 357 ms
- Receiving the response
- 27 ms
- Total
- 395 ms
- Response opened
- 372 ms
- First visible word
- 372 ms
- Answer complete
- 395 ms
Tokyo468 ms
Tokyo- Finding the server
- 1 ms
- Reaching the server
- 2 ms
- Setting up security
- 9 ms
- Waiting for the server: most of the time
- 435 ms
- Receiving the response
- 22 ms
- Total
- 468 ms
- Response opened
- 446 ms
- First visible word
- 447 ms
- Answer complete
- 468 ms
The phases come from the controlled run of 13 September 2026. Typical times pool daily requests; slower times use controlled full runs. 5 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo.
Finding the server is measured once per location, so it sits outside these bars. Open a city to see it.
Bars show the typical request, so the parts add up to its total. Why
Mistral still feels slow?
- Check Mistral’s status page
- Elevated latency and error rates are posted there. A slow first token during an incident is not your region.
- Short everywhere, a little longer far from Europe
- Amsterdam waited 239 ms in the controlled run; Montreal about 60 ms more, San Francisco and Singapore about 120 ms more, Tokyo about 200 ms more. The single Cloudflare address does not say where the model runs, but the pattern suggests the origin is in Europe. For users in Asia, most of the extra fifth of a second is distance.
- What this does not measure
- Le Chat, tokens per second, long prompts, other Mistral models including the Magistral reasoning models, and the Batch API. The ChatGPT, Claude and Grok figures on this page use the same prompt on the same days at low reasoning effort, which adds a short thinking phase Mistral Medium does not have.
Response time over the last 30 days
Show the numbers
| Day | Amsterdam | San Francisco | Montreal | Singapore | Tokyo | Mumbai |
|---|---|---|---|---|---|---|
| 27 August 2026 | 247 ms | 403 ms | 339 ms | 522 ms | 463 ms | — |
| 28 August 2026 | 232 ms | 369 ms | 290 ms | 504 ms | 461 ms | — |
| 29 August 2026 | 294 ms | 358 ms | 316 ms | 417 ms | 446 ms | — |
| 8 September 2026 | 255 ms | 416 ms | 403 ms | 372 ms | 452 ms | 355 ms |
| 9 September 2026 | 236 ms | 355 ms | 325 ms | 402 ms | 422 ms | 319 ms |
| 10 September 2026 | 247 ms | 320 ms | 295 ms | 347 ms | 423 ms | 304 ms |
| 11 September 2026 | 248 ms | 357 ms | 310 ms | 415 ms | 460 ms | 366 ms |
| 12 September 2026 | 200 ms | 338 ms | 279 ms | 373 ms | 410 ms | 324 ms |
| 13 September 2026 | 245 ms | 332 ms | 258 ms | 400 ms | 414 ms | 322 ms |
| 14 September 2026 | 262 ms | 346 ms | 329 ms | 375 ms | 486 ms | 354 ms |
| 15 September 2026 | 220 ms | 361 ms | 552 ms | 361 ms | 455 ms | 312 ms |
| 16 September 2026 | 227 ms | 373 ms | 317 ms | 343 ms | 429 ms | 295 ms |
About this measurement
POST api.mistral.ai/v1/chat/completions · 45 requests per location over 9 daily runs · 8 September 2026 to 16 September 2026
What we tested · AI response. Time to first visible token from mistral-medium-latest on a one-word prompt: streaming, 256 max tokens, no reasoning setting, paid key.
How LatencyRadar measures response time →
Technical details
Doesn’t measure: Le Chat, tokens per second, long prompts, or other Mistral models.
One Mistral surface: time to first visible token from mistral-medium-latest at api.mistral.ai/v1/chat/completions, a one-word prompt, streaming, 256 max tokens, paid key. The request sets no reasoning option because Mistral Medium answers directly; there is no thinking phase to include or leave out. Mistral's API uses the OpenAI-compatible event format, so the same parser that reads ChatGPT's stream reads this one.
Not measured: Le Chat, tokens per second, long prompts, other Mistral models including the Magistral reasoning models, or the Batch API. The ChatGPT, Claude and Grok figures on this page are the same prompt on the same days at low reasoning effort. Their daily medians ran 0.7 to 2.0 seconds (ChatGPT), 1.2 to 2.3 seconds (Claude) and 1.1 to 3.5 seconds (Grok) while Mistral's ran 0.20 to 0.55 seconds. Part of that gap is thinking Mistral Medium never does, so the two sets of numbers describe different requests and are not a ranking.
The published measurements cover 8 to 16 September 2026, the first nine daily runs since collection restarted on 8 September. Typical first-token times are 245 ms in Amsterdam to 442 ms in Tokyo, with Montreal at 310 ms, San Francisco at 347 ms and Singapore at 373 ms. Daily medians moved inside a band of 60 to 100 ms in four cities. Montreal had one slower day, 552 ms on 15 September against 258 to 329 ms on the other eight. In the controlled run the slower responses sat furthest above the typical in San Francisco (747 ms) and Singapore (842 ms), so budget for the tail there even though the typical wait is short.
| Test location | Typical response time | Slower response time | Requests |
|---|---|---|---|
| Amsterdam | 245 ms | 364 ms | 20 |
| San Francisco | 347 ms | 747 ms | 20 |
| Montreal | 310 ms | 381 ms | 20 |
| Singapore | 373 ms | 842 ms | 20 |
| Tokyo | 442 ms | 570 ms | 20 |
| Mumbai | Collecting since 8 Sept 2026 | ||
Typical: half of the requests finished within this time (technical: p50 of time to first visible token). Slower: 95% of requests finished within this time (technical: p95). Statistics
- Request
- Authenticated streaming POST to api.mistral.ai/v1/chat/completions
- Measured from
- Amsterdam · San Francisco · Montreal · Singapore · Tokyo
- Requests
- Daily measurements from 8 September 2026 to 16 September 2026. 5 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo. · 8 September 2026 to 16 September 2026
- Model
- mistral-medium-latest · no reasoning option (the model answers directly) · max_tokens 256 · default temperature
- Prompt
- "Say 'ok' and nothing else." Identical across every city and every LLM API we benchmark
- Comparison figures
- ChatGPT (gpt-5.6-sol), Claude (claude-opus-5) and Grok (grok-4.6), all at low reasoning effort, from the same daily collection between 8 and 16 September 2026. Those three think briefly before the first token; Mistral Medium does not
- Timings taken
- Parsed from the token stream: stream open, first visible token, completion, plus DNS, connect and TLS
- Report coverage
- 5 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo.
- Daily requests in each location
- Amsterdam: 45 · San Francisco: 45 · Montreal: 45 · Singapore: 45 · Tokyo: 45
- Typical response time
- Median of the eligible daily requests in each included location.
- Slower response time
- Median of recent controlled full-run p95 values in each included location.
Mistral API
https://api.mistral.ai/v1/chat/completions
Independent measurement by LatencyRadar. Not affiliated with Mistral AI. · How we measure (v1.1)
How fast is your API compared to Mistral's?
Run a free speed test from multiple cities and find out where your users are waiting. No setup, no account required.
No account required · Takes about 30 seconds.
