A disambiguation, then a distinction. The Kimi in this Mistral vs Kimi for trading comparison is the model family from Moonshot AI (no relation to the similarly named chat assistant apps) — and the distinction is about what its win means: of the pair's 2 head-to-head results, Kimi's is the only one that doubled as a season championship. The record runs Seasons 5–6, the only completed TradeRank seasons Mistral Medium 3.5 and the Moonshot slot have both traded (Kimi ran K2.6, then K2.7 Code); identical per-season conditions — the same 10-asset menu, the same daily clock, matching $10,000 simulated stakes — and every figure below is lifted from the generated evidence pack linked at the end, not from any model's recall.
Which versions traded each season
| Season | Dates | Mistral version | Kimi version | Asset universe | Field |
|---|---|---|---|---|---|
| Season 5 | May–Jun 2026 | Mistral Medium 3.5 | Kimi K2.6 | 10 crypto assets | 10 models |
| Season 6 | Jun–Jul 2026 | Mistral Medium 3.5 | Kimi K2.7 Code | 10 crypto assets | 11 models |
Head-to-head results by season
| Season | Mistral return | Kimi return | Gap (Mistral − Kimi, pts) | Rank (Mistral / Kimi) | Trades (Mistral / Kimi) | Win rate (Mistral / Kimi) | Max drawdown (Mistral / Kimi) | Winner |
|---|---|---|---|---|---|---|---|---|
| Season 5 | +9.55% | +5.78% | +3.77 | 3rd of 10 / 4th of 10 | 6 / 14 | 83.3% / 50% | 7.94% / 10.60% | Mistral |
| Season 6 | -2.95% | +2.14% | -5.67 | 9th of 11 / 1st of 11 | 8 / 13 | 25% / 30.8% | 6.24% / 7.36% | Kimi |
Returns, season by season
What Separates the Mistral vs Kimi for Trading Wins
Head-to-head cells treat every win identically; fields do not. Mistral's Season 5 win was a contest between neighbors — 3rd against 4th of 10, both books green, settled by +3.77 points with the field's top spots out of reach of both. Kimi's Season 6 win had no one above it at all: +2.14% official led all 11 models, in a season where +2.14% was the best return any of the 11 models produced. Mistral, meanwhile, was 9th — so the same head-to-head column that recorded a tight loss for Kimi recorded a 5.67-point, top-versus-bottom-third result for Mistral.
Both readings are the same data at different magnifications. On the pairwise scoreboard this is simply 1-1. On the field's scoreboard, one of these models has a season title in the shared record and the other's best result is a podium place — with the caveat that ranks restate the returns rather than adding to them, and that each claim spans a single season.
Each season's return against its deepest drawdown
The Champion Took the Deeper Dips
Kimi's maximum drawdown exceeded Mistral's in both seasons — 10.60% to 7.94%, then 7.36% to 6.24% — and the outcomes attached to that ordering could hardly differ more: a head-to-head loss from 4th place, then a field championship. Whatever Season 6's title was built on, it was not a shallower worst moment; the champion dipped further than the 9th-place book it beat.
The column's limits are the usual ones, tightened by the sample. 2 seasons of a fixed ordering, with opposite results attached, supports no safety story in either direction; maximum drawdown is one worst moment on an equity path whose destination the return already reports; and the pack carries no volatility measure to deepen the picture. What the cells document — precisely and only — is that in this pair, the steadier-looking equity path never belonged to the season winner's opponent twice.
Trade count by season
The Openings the Snapshots Could Attribute
Season 5's early cycles supplied both models' first attributable outcomes, and the entries read differently in each log. Kimi's (Kimi K2.6) earliest attributable gain was a ZEC long taken a day in, its log reasoning that "strong weekly breakout momentum supports a long position despite elevated RSI". Mistral's was its TRX long from the opening cycle, logged as the day's only pass through its entry rules. Kimi's earliest attributable loss was an ETH short from the opening window; Mistral's, a DOGE short in mid-June.
The pack reconstructs these from position-state changes between consecutive daily equity snapshots — no fills, sizes or exits — and keeps only each model's earliest attributable gain and loss. 4 cells of decision evidence describe the record's earliest attributable moments, nothing after them; the title and the podium above were both earned in the parts of the record the snapshots cannot individually attribute.
How We Measured the Championship Season
Season 6's championship framing makes the measurement conventions worth spelling out precisely. The title is an official-standings fact: +2.14% led the 11-model field after the season's forced liquidation settled every book. The pair's gap for that season, though, is computed on the pack's comparability basis — each model's last pre-liquidation snapshot, +2.64% for Kimi against -3.02% for Mistral — so that it can sit beside Season 5's retained-exposure gap without mixing conventions. Winner and title are identical on either basis.
Inside each season the conditions were standard: a shared 10-asset crypto list, one decision per day, live prices into simulated $10,000 accounts, 0.1% modeled fees, no slippage or borrowing modeled. Across seasons, the field grew to 11, Kimi's build moved from K2.6 to K2.7 Code, the prompt regime shifted mid-Season 6 to the medium-term investor framing, with an era note marking a universe expansion toward US equities whose timing is unprovable. Every figure was produced by a deterministic generator that stores the paired record in an evidence pack under a content hash; the built page is screened against that pack before it publishes, and the writing model supplied no numbers.
Limitations: One Title, Two Observations
The title is real and the sample is 2 — hold both. Kimi's Season 6 championship is the strongest single-season fact in this pair's record, and it is still a single season, won by a build (K2.7 Code) that did not trade the other one; 'Kimi finished 1st' and 'Kimi is the better trader' remain different sentences. Mistral's Season 5 win is a single season too, achieved against the other Kimi build. Season 5's returns were mostly open marks on both sides — Kimi's +5.78% sat over a realized -$478.43 — while Season 6's forced liquidation made its books mostly settled, so the 2 seasons' headline returns are not even the same kind of number without the comparability adjustment. Win rates count open positions throughout. And the pair's next season could flip any sentence above: that is what a 2-observation record means. The running version of this matchup lives on the live LLM trading benchmark; the closed version, every figure of it, sits in the Mistral vs Kimi evidence pack.