Research benchmark only. Not financial advice. Crypto trading can lose all capital.

AI Trading Ethereum: 26 LLMs, 385 Trades, 5 Seasons

Ethereum has been tradeable since the arena opened, so its record runs the full length of the archive. Most competitors have held ether at some point, argued with themselves about it in their reasoning, and eventually written down why they let go.

MiniMax M3 holds the strongest aggregate realized result on Ethereum: $965.73 across 3 closed trades. That is one asset's record, not an overall ranking — 26 models have traded an ETH position here, over 5 seasons whose trade log survives.

  • 385ETH trades on record (floor)
  • 184closed with a realized result
  • $590.67realized, all models combined
  • 12open positions in the live season

Best LLM for Trading Ethereum

The leading ETH results rest on a handful of closed legs each, and the biggest gains among them came late in the archive. Further down the full table, the first row from the early seasons closed many more legs for a smaller total. Same column, same units, very different amounts of evidence, which is why every entry states how many legs it closed.

  1. MiniMax M3$965.73 realized on ETH

    3 closed legs of 7 trades, in Seasons 6 and 8.

  2. Kimi K3$630.72 realized on ETH

    3 closed legs of 6 trades, in Season 8.

  3. Inkling$320.27 realized on ETH

    1 closed leg of 2 trades, in Season 8.

Ranked by aggregate realized profit and loss on ETH alone, over the seasons whose trade log survives. Equal results are joint, never ordered. Open positions carry no realized figure and are excluded.

Can AI Trade Ethereum?

They already do, and an ETH position is ordinary business: opened on a stated thesis, sized as a share of equity, closed when the thesis breaks or the season ends. What varies is the justification, and that is quoted in full further down.

26 model versions have executed an ETH trade in TradeRank, and 39 versions have traded it or written about it in their decisions. The realized total across every closed leg is $590.67, and that total is a floor rather than a count.

Every Model's ETH Record

One row per model version. A cell reads n/a where the answer is unknown: the trade log for those seasons did not survive. A known zero is shown as a zero instead — a row that traded and closed nothing realized nothing, which is a measurement rather than a gap.

ModelProviderTradesClosedRealized P&LSeasons
MiniMax M3MiniMax73$965.73Seasons 6 and 8
Kimi K3Moonshot63$630.72Season 8
InklingThinking Machines21$320.27Season 8
Gemini 3 ProGoogle2412$265.88Seasons 1 and 2
Claude Fable 5Anthropic53$258.74Seasons 6 and 8
Nemotron 3 UltraNVIDIA63$220.64Seasons 6 and 8
ox-alphaStealth31$77.35Season 8
Mistral Medium 3.5Mistral AI42$68.78Seasons 6 and 8
Grok 4-1 FastxAI3720$47.52Seasons 0 and 1
Qwen3.8 2.4T A95BAlibaba31$39.12Season 8
Claude Sonnet 4.5Anthropic21$12.19Season 1
Claude Opus 4.8Anthropic20$0.00Season 6
GLM-5.2Zhipu AI42-$23.13Season 6
GPT-5.5OpenAI21-$24.01Season 6
Kimi K2.7 CodeMoonshot42-$26.92Season 6
Qwen 3.7 PlusAlibaba21-$35.70Season 6
Claude Haiku 4.5Anthropic5626-$70.28Seasons 0 and 1
Gemini 3.5 FlashGoogle21-$72.97Season 6
Gemini 2.5 FlashGoogle2010-$80.60Seasons 0 and 1
DeepSeek V4 ProDeepSeek62-$83.13Season 6
Grok 4.3xAI21-$126.74Season 6
o4-miniOpenAI2010-$126.90Season 1
Qwen3Alibaba4724-$215.63Season 1
GPT-5 MiniOpenAI4518-$247.53Seasons 0 and 1
Kimi K2Moonshot3015-$581.34Season 1
DeepSeek R1DeepSeek4421-$601.39Season 1
Claude Opus 4.6Anthropicn/an/an/aSeason 3
Claude Opus 4.7Anthropicn/an/an/aSeasons 4 and 5
DeepSeek V3.2DeepSeekn/an/an/aSeason 3
Gemini 3.1 ProGooglen/an/an/aSeasons 3 and 4
GLM-5.1Zhipu AIn/an/an/aSeasons 4 and 5
Grok 4.20xAIn/an/an/aSeason 3
Grok 4.20 MAxAIn/an/an/aSeason 4
Kimi K2.5Moonshotn/an/an/aSeason 3
Kimi K2.6Moonshotn/an/an/aSeasons 4 and 5
MiniMax M2.5MiniMaxn/an/an/aSeason 3
MiniMax M2.7MiniMaxn/an/an/aSeason 4
Qwen 3.5 PlusAlibaban/an/an/aSeason 3
Qwen 3.6 PlusAlibaban/an/an/aSeasons 4 and 5

Ethereum AI Trading Performance by Season

Ethereum concentrates most of its executed volume in one early season and then thins out sharply. The middle of the archive kept models writing about ether every cycle without leaving a single recoverable fill behind.

Trade counts are not comparable across the whole column: the early seasons ran several decision cycles a day and the later ones settle everything at one fixed hour, so the Cycle column states which regime each row belongs to.

SeasonCycleTradesClosedRealized P&LModels trading
Season 0intraday2815$95.274
Season 1intraday295141-$1,687.7210
Season 2intraday21-$5.631
Season 3once dailyn/an/an/an/a
Season 4once dailyn/an/an/an/a
Season 5once dailyn/an/an/an/a
Season 6once daily3717-$766.7612
Season 7once dailyn/an/an/an/a
Season 8once daily2310$2,955.518

A cell reads n/a where the answer is unknown: the trade log for those seasons did not survive. A known zero is shown as a zero instead — a row that traded and closed nothing realized nothing, which is a measurement rather than a gap.

Who Holds ETH Right Now

The season now running, as of 2026-09-12. Every table above this block is frozen archive; this block and the fourth figure in the hero are the only parts that move.

ModelSideEntryLast price
Claude Fable 5.1long$2,535.62$2,523.42
DeepSeek V4.1 Flashlong$2,535.62$2,523.42
Gemini 3.8 Flashlong$2,535.62$2,523.42
GLM-5.3long$2,535.62$2,523.42
Kimi K3long$2,535.62$2,523.42
LongCat 2.0long$2,535.62$2,523.42
MiniMax M3long$2,535.62$2,523.42
Muse Spark 1.3long$2,535.62$2,523.42
Nemotron 3 Ultralong$2,535.62$2,523.42
Nightjarlong$2,535.62$2,523.42
Qwen3.8 Max 0902long$2,535.62$2,523.42
Seed 2.1 Turbolong$2,535.62$2,523.42

Season so far: 12 trades, 0 closed legs, realized n/a. The live leaderboard carries the current standings.

ETH is flat versus entry and remains in a durable uptrend with higher lows and BTC structure intact, so the long thesis stands and the 2340 thesis-failure level remains armed.

Nightjar, its own decision reasoning (unedited), Season 9

ETH's medium-term uptrend thesis remains intact; price is consolidating above $2500 support after the recent $2666 swing high, and the $2380 invalidation level still validates the trend structure.

Seed 2.1 Turbo, its own decision reasoning (unedited), Season 9

ETH is still holding above the breakout consolidation near 2400, so the medium-term momentum thesis remains intact; a break below that zone would weaken the setup.

Qwen3.8 Max 0902, its own decision reasoning (unedited), Season 9

The Best and Worst ETH Calls

Selected deterministically from the trade log: the strongest and weakest closed ETH legs that carry a written rationale, whichever model made them. A larger leg with no reasoning attached to it is not shown here, because the section exists to publish the reasoning.

Largest quoted ETH leg: $1,124.22, MiniMax M3

Original target of $2,200 was reached; now at $2,326 with 4h RSI 94.62 signaling extreme overbought. Full exit locks in +22.6% gain and frees capital to redeploy at better risk/reward after a pullback to retest breakout…

MiniMax M3, its own trade reasoning (unedited apart from length), Season 8

Weakest quoted ETH leg: -$144.73, Mistral Medium 3.5

ETH short under significant pressure (-5.7% PnL) with weekly bearish score only 15/100. Trend alignment weak despite daily being neutral. Cut losses to preserve capital.

Mistral Medium 3.5, its own trade reasoning (unedited), Season 6

What the Models Said About Ethereum

Every quote below is a model's own reasoning for an ETH decision, unedited apart from length. Selection is deterministic and recorded in the pack.

Original target of $2,200 was reached; now at $2,326 with 4h RSI 94.62 signaling extreme overbought. Full exit locks in +22.6% gain and frees capital to redeploy at better risk/reward after a pullback to retest breakout…

MiniMax M3, its own trade reasoning (unedited apart from length), Season 8

ETH reached its rebound target with sharply overbought RSI, so I am locking in profits per the thesis-completion rule.

Inkling, its own trade reasoning (unedited), Season 8

Banking half the +23% gain with price well through the 2,050-2,180 thesis target; keeping a runner for the neutral-funding regime shift.

Kimi K3, its own trade reasoning (unedited), Season 8

Original oversold bounce thesis invalidated by price action: ETH weekly RSI now above 50 (53.11) and 4h RSI overbought (87.88). Locking in +10.8% profit.

Mistral Medium 3.5, its own trade reasoning (unedited), Season 8

Target $2100-2200 reached; weekly RSI normalized from oversold; mean-reversion thesis complete.

Nemotron 3 Ultra, its own trade reasoning (unedited), Season 8

Buy the confirmed range breakout in the highest-liquidity crypto with uncrowded positioning and room on the weekly.

Claude Fable 5, its own trade reasoning (unedited), Season 8

A small position adds exposure to the strongest tradeable crypto setup without overcommitting to an overbought tape.

Qwen3.8 2.4T A95B, its own trade reasoning (unedited), Season 8

Deploying the previously telegraphed ETH trigger — a healthy cool-off that held the breakout base — at measured size with the invalidation placed where the flag is objectively dead.

ox-alpha, its own trade reasoning (unedited), Season 8

ETH offers the strongest technical improvement among liquid tradeables, allowing a patient, well-defined long with clear invalidation criteria.

Inkling, its own decision reasoning (unedited), Season 7

Buying the strongest liquid major on a low-volume pullback to breakout support, sized at 30% of equity with a ~5% structural invalidation.

Kimi K3, its own decision reasoning (unedited), Season 7

Best risk/reward among tradeable assets: oversold weekly, improving daily momentum, deep liquidity, and strong fundamental tailwinds.

Nemotron 3 Ultra, its own decision reasoning (unedited), Season 7

Highest relative strength in the tradeable universe with a fresh, volume-confirmed breakout holding on the retest.

Claude Fable 5, its own decision reasoning (unedited), Season 7

ETH demonstrates relative strength with oversold weekly RSI and bullish daily momentum divergence.

Mistral Medium 3.5, its own decision reasoning (unedited), Season 7

ETH short under significant pressure (-5.7% PnL) with weekly bearish score only 15/100. Trend alignment weak despite daily being neutral. Cut losses to preserve capital.

Mistral Medium 3.5, its own trade reasoning (unedited), Season 6

Daily EMA-26 shows 0 direction; position at -8% loss with weak overall score (15)

Grok 4.3, its own trade reasoning (unedited), Season 6

Daily trend reversed — price reclaimed the daily EMA-26 (1d score now neutral) and broke above the two-week range on strong volume. Composite fell to 15/100, the weakest of all holdings. Cutting the -4.8% loss before…

Claude Fable 5, its own trade reasoning (unedited apart from length), Season 6

ETH daily EMA-26 is now neutral (0 weight), breaking the required daily agreement for short entries. Price moved against us by 4%. Close to preserve capital.

DeepSeek V4 Pro, its own trade reasoning (unedited), Season 6

Daily EMA-26 score is 0 (no bearish agreement), composite 15 is far below the 50 threshold, making the short invalid; closing limits further loss.

Kimi K2.7 Code, its own trade reasoning (unedited), Season 6

Daily trend has flipped to bullish and held, violating the weekly/daily bearish alignment necessary to sustain the short.

Gemini 3.5 Flash, its own trade reasoning (unedited), Season 6

Daily trend reversed bullish for 2+ candles (1700→1731). Weekly/daily now disagree. Exit rule triggered—cut the -4.3% loss.

GLM-5.2, its own trade reasoning (unedited), Season 6

Daily trend reversed bullish (▲25) against short position while weekly remains bearish; held for multiple daily candles — exit rule triggered

Nemotron 3 Ultra, its own trade reasoning (unedited), Season 6

Daily trend has reversed to bullish (1d:▲) while weekly remains bearish (1w:▼). Recent daily closes (1700.57, 1731.24) held above daily EMA-26 for 2+ candles, triggering exit rule. Strategy mandates staying flat when…

MiniMax M3, its own trade reasoning (unedited apart from length), Season 6

Daily trend reversing up for 2+ candles against the short; composite score < 50 mandates staying flat.

Qwen 3.7 Plus, its own trade reasoning (unedited), Season 6

ETH weekly remains bearish, but the daily trend is now neutral and 4h momentum is bullish with elevated short-term RSI, so the short no longer has clean weekly/daily downside alignment.

GPT-5.5, its own trade reasoning (unedited), Season 6

What This Page Does and Does Not Measure

  • Trade records cover Seasons 0, 1, 2, 6 and 8. Seasons 3, 4, 5 and 7 ran Ethereum but their trade logs are unrecoverable, so those rows read n/a.
  • Season 1 survives only as a partial log, so every all-time total on this page is a floor rather than a count.
  • Decision reasoning covers Seasons 1, 3, 4, 5, 6, 7 and 8, which is why a model can be quoted in a season with no trades beside it.
  • 33.2% of the 184 closed ETH legs finished positive. This is a closed-trade rate on this asset only. It is not the standings win rate, which counts open positions too, and the two are not comparable in either direction.
  • Everything here is paper trading. Each model starts from the same simulated balance, pays a fee on every fill, and risks no capital. How the competition works.

Ethereum AI Trading — FAQ

What is the best LLM for trading Ethereum?

MiniMax M3 holds the strongest aggregate realized result on ETH: $965.73 across 3 closed trades. That is one asset's record, not an overall ranking, and it covers only the seasons whose trade log survives.

Can AI trade Ethereum?

They already do. 26 models have executed 385 ETH trades in TradeRank, 184 of them closed with a realized result.

How many AI models have traded ETH?

26 model versions have executed an ETH trade. A wider set — 39 versions — traded it or wrote about it in their decisions, including seasons whose trade log did not survive.

Did the models make or lose money on Ethereum?

Across the closed ETH trades on record the realized total is $590.67. Season 1 survives only as a partial log, so that is a floor. Open positions carry no realized figure and are not in that total.

Which seasons does this Ethereum record cover?

Trades: Seasons 0, 1, 2, 6 and 8. Seasons 3, 4, 5 and 7 ran the asset but their trade logs are unrecoverable, so those rows read n/a rather than zero.

Are the Ethereum results on this page live trades?

No. TradeRank is a paper-trading benchmark: each model starts from the same simulated balance, pays fees on every fill, and risks no capital.