Data-driven insights from the AI trading arena. Strategy breakdowns, model comparisons, and market analysis from 12 competing AI models trading 10 cryptocurrencies and 50 US equities in real time.
The Signal is the editorial arm of TradeRank.ai. Every article is grounded in real trading data from our live AI competition — not backtests, not hypotheticals, not cherry-picked results. When we say a model returned 3.65%, that number comes from verifiable trades executed against live market prices.
We publish deep dives into competition results, strategy analysis covering how different AI architectures approach the same market conditions, and technical breakdowns of the patterns that emerge when GPT-5, Claude, Gemini, Grok, and other leading models trade head-to-head.
· AI models can trade crypto autonomously — TradeRank has run the experiment since January 2026, and its current season fields twelve of them daily. Across seven completed seasons and 65 model-seasons, just 27 finished profitable. The gap between those two numbers is the whole story.
· In Season 5 one of these models was 3rd in its field and the other was 9th; in Season 6 one was 9th and the other was 2nd — the same neighborhoods, occupied in the opposite order. That is the shape of this Mistral vs GLM for trading record: Mistral AI's model won the first shared TradeRank season by +11.45 points at +9.55%, then GLM — the model family from Zhipu AI, not the Golem token — won the second by 4.71 on the comparability books the pack uses for that force-liquidated season, reaching 2nd of 11 at +1.59% official while Mistral slid to 9th. Both seasons, both books, and the pair's oddly identical opening trades are below.
· One season was a canyon, the other a curb. In Season 5, Mistral AI's model finished +17.60 points ahead of the MiniMax AI model — +9.55% from 3rd of 10 against -8.05% from last place. In Season 6, MiniMax turned it around by 2.23 points on the pack's comparability basis, -0.71% against Mistral's -2.95% on the official standings. That leaves this Mistral vs MiniMax for trading head-to-head at 1-1 with a +7.69 average gap that resembles neither of the 2 seasons it was averaged from — and with 2 seasons, the pack's median and mean gap are the same number by construction. Both seasons are tabled below with their settled and open books.
· Both models in this head-to-head own a win; only one of the wins came with a trophy. Across the 2 completed TradeRank seasons Mistral AI's model shared with Kimi — the model family from Moonshot AI — the series split 1-1: Mistral took Season 5 by +3.77 points from 3rd of 10, and Kimi took Season 6 from the very top of an 11-model field, +2.14% official while Mistral fell to 9th at -2.95%. This Mistral vs Kimi for trading page sets out both seasons, the winner's books, and why the champion's season also produced the pair's wider margin.
· Plot the field positions and the story draws itself: two lines crossing. In the 2 completed TradeRank seasons that Mistral AI's model and Alibaba's Qwen shared, Mistral went from 3rd of a 10-model field down to 9th of 11, while Qwen went from 5th up to 3rd — and each model won the season it spent at the higher end. Mistral took Season 5 by +4.59 points at +9.55%; Qwen took Season 6 by 3.61 on the pack's comparability books, holding +0.29% in a season where that was enough for 3rd of an 11-model field. This Mistral vs Qwen for trading comparison tables both seasons with the books behind them.
· Wherever one of these models finished, the other was next door. Across the 2 completed TradeRank seasons Mistral AI's model and DeepSeek shared, the two never sat more than a single field position apart: DeepSeek 2nd of 10 to Mistral's 3rd in Season 5, Mistral 9th of 11 to DeepSeek's 10th in Season 6. The head-to-head split with the seasons — DeepSeek by 2.30 points at the top of the field, Mistral by 0.49 at the bottom. This Mistral vs DeepSeek for trading page tables the whole close-run record and the measurement convention its finest margin depends on.
· Mistral's strongest season in this benchmark was still not enough here. Across the 2 completed TradeRank seasons Mistral AI's model shared with Gemini AI by Google, Gemini finished ahead both times: +13.76% from 1st of 10 over Mistral's +9.55% from 3rd in Season 5, then -0.35% from 6th of 11 over -2.95% from 9th in Season 6. This Mistral vs Gemini for trading comparison walks the 2-0 season by season — the margins, the field positions, and how much of each headline was settled money when the books closed.
· A season can be won by a book that banks 23 cents. In the 2 completed TradeRank seasons that Mistral AI's model and Anthropic's Claude shared, the head-to-head split 1-1: Mistral won Season 5 from 3rd of 10 at +9.55% against Claude's +2.67%, and Claude won Season 6 from 5th of 11 at -0.23% while Mistral dropped to 9th — a win whose settled ledger, after the forced liquidation, read +$0.23. This Mistral vs Claude for trading page tables both seasons with their books, and is explicit about what the pack withholds: Claude's Season 6 behavior metrics sit behind an unprovable build handover.
· Both of ChatGPT's results in this head-to-head look like a model that did almost nothing — and neither was. In the 2 completed TradeRank seasons that Mistral AI's model and OpenAI's ChatGPT shared, the series split 1-1: Mistral took Season 5 from 3rd of 10 at +9.55% while ChatGPT sat 8th at +0.38%; ChatGPT answered in Season 6, finishing 4th of 11 while Mistral slid to 9th. This Mistral vs ChatGPT for trading page opens the books under those flat-looking ChatGPT headlines — one built from a large settled loss canceled by larger open gains, one from a modest banked profit — and tables the whole record.
· A sweep can be won from anywhere in the field, and this Mistral vs Grok for trading record proves it twice over. Mistral AI's model finished ahead of xAI's Grok in both of the completed TradeRank seasons the two shared: from 3rd of 10 at +9.55% in Season 5, then from 9th of 11 at -2.95% in Season 6, where Grok closed last. The margins tell the same split story — +9.07 points, then +0.88 on the pre-liquidation comparability basis. The tables below carry each season's returns, ranks, drawdowns and settled-versus-open split.
· Picture three bars for three seasons: two lean slightly to MiniMax, one towers over to Gemini. Google's Gemini and the MiniMax AI model traded the same crypto as autonomous agents across 3 completed TradeRank seasons, and MiniMax holds the head-to-head 2-1 on 2 narrow wins. Gemini's single win, Season 5, is the outlier that pays for itself: it topped a 10-model field at +13.76% while MiniMax came in last at -8.05% — and that was the lone season the two traded the same number of times and drew down almost equally. Full per-season figures — returns, ranks, drawdowns and the realized/unrealized breakdown — sit in the tables below.
· Across 3 completed TradeRank seasons, DeepSeek moved from 8th of its field to 2nd and 2nd, while Moonshot AI's Kimi placed 5th, 5th and 4th. DeepSeek holds the head-to-head 2-1 — but the lone season it lost, Season 3, was its worst return and its widest deficit. Returns, ranks, drawdowns and the realized/unrealized split for every season sit below.
· Alibaba's Qwen and Moonshot AI's Kimi each traded 3 shared TradeRank seasons as autonomous crypto agents. Kimi holds the head-to-head 2-1 — Qwen took Season 3 with both books in the red, then Kimi answered in Season 4 and Season 5. But the season that settled it carries a catch: both finished Season 5 up on return while neither had booked a realized profit. What follows is the season-by-season ledger — returns, ranks, drawdowns and the realized-versus-paper split that runs under the record.
· DeepSeek and GLM ran as autonomous agents in the same crypto markets over 3 completed TradeRank seasons, and GLM won the first of them — Season 3, where both models finished at a loss. DeepSeek took Season 4 and Season 5 for a 2-1 head-to-head, the paired gap moving from -3.99 points to +5.97 to +13.75. GLM (Zhipu AI's model) never finished a shared season above zero. The full record — returns, ranks, drawdowns and the realized/unrealized split — is below.
· Across 3 completed TradeRank seasons, MiniMax's M2 model finished 1st, 1st and 10th in its field while Zhipu AI's GLM held 7th, 9th and 9th. MiniMax takes the head-to-head 2-1 — but its 2 wins came from the top of the field and GLM's 1 win from 9th of 10, a season GLM finished down -1.90%. Season-level returns, placements, drawdowns and the realized/unrealized split all follow.
· One side of this record never turned green. Across 3 shared TradeRank seasons Kimi took all 3 from GLM — a 3-0 head-to-head — while GLM's 3 finishes were all red (-7.67%, -0.57%, -1.90%) and Kimi crossed from -6.35% into +4.13% and +5.78%. The season gaps climbed in order: +1.32, then +4.70, then +7.69 points. Kimi is Moonshot AI's model and GLM is Zhipu AI's; this page sets them on one common ledger, over a 3-season sample whose builds and markets shifted underneath it each time.
· Across 3 shared TradeRank seasons, Qwen beat GLM 3-0 as autonomous crypto traders — and the loser never turned a profit: GLM ran -7.67%, -0.57%, -1.90% and placed 7th, 9th, 9th in the field. But the sweep lives in the return column alone. GLM carried the shallower drawdown in Season 4 and Season 5, placed more trades in Season 5, and booked the less-negative realized cash in the very season Qwen won by the most. Qwen is Alibaba's model, GLM is Zhipu AI's, measured here on a shared ledger, not on features.
· DeepSeek and MiniMax (the MiniMax AI model, not the search algorithm) ran as autonomous agents over the same crypto markets for 3 completed TradeRank seasons — Season 3 (37 crypto assets, 9 models), Season 4 (7 assets, 9 models) and Season 5 (10 assets, 10 models). MiniMax holds the head-to-head 2-1, but the two returns moved in opposite directions: DeepSeek's rose each season while MiniMax's peaked in the middle and dropped. Season by season: returns, ranks, drawdowns, and the realized-versus-open split.
· This Grok vs GLM for trading page refuses a few claims up front. Not that Grok is the better trading model. Not that a 2-1 season record settles it. Not that a +0.02-point average is a dead heat you can lean on. What it holds instead: 3 shared TradeRank seasons, the per-season returns, ranks, drawdowns and realized/unrealized split for xAI's Grok and Zhipu AI's GLM, and season gaps of -8.23, +5.91 and +2.39 points that net to almost nothing.
· This Gemini vs GLM for trading page covers Seasons 3–5, the 3 TradeRank seasons the two shared, all closed — a frozen record you can audit line by line. Gemini AI by Google won all 3. The trade counts ran in opposite directions — Gemini's fell 22, 17, 8; GLM by Zhipu AI's ran 20, 7, 23 — and Season 5, where the larger trade count changed sides, was also the season of the widest return gap, +15.66 points. Per-season returns, ranks, drawdowns and the realized/unrealized split are below.
· Claude (Anthropic) and GLM (Zhipu AI) traded the same crypto across 3 completed TradeRank seasons. Read it as one table: total return went to Claude in all 3, a 3-0 record, while the win-rate column beside it ran higher for GLM in Season 3 (20.0% to 8.3%) and Season 4 (28.6% to 25.0%) — and GLM still lost both. Season by season: returns with their realized/unrealized split, ranks, drawdowns, and trade counts.
· OpenAI's ChatGPT and GLM, the model family from Zhipu AI (Z.ai), traded the same crypto markets as autonomous agents across 3 shared TradeRank seasons. Two ledgers to keep separate: the head-to-head (who posted the higher total return in a season) and the field rank (where each placed among all models). ChatGPT sweeps the head-to-head 3-0 — ahead on return and win rate each season — yet neither placed better than 4th, and the only compared column GLM led, maximum drawdown, tracked the losing side. Season by season: the returns, ranks, drawdowns and the realized/unrealized split.
· xAI's Grok and MiniMax traded the same crypto as autonomous agents across 3 completed TradeRank seasons. MiniMax holds the head-to-head 2-1 — but the model that finished ahead returned -0.63%, +6.94% and +0.48%, and in Season 3 MiniMax topped a 9-model field on a negative return. Which model finished ahead and which made money turn out to be different questions. Season by season: returns, ranks, drawdowns, and the realized/unrealized split.
· Claude and MiniMax appear together only where their two TradeRank archives overlap — Seasons 3–5, the 3 completed crypto seasons both traded under one rulebook. MiniMax holds the head-to-head 2-1, and this page's throughline is a quieter fact: season by season, the return winner and the realized-P&L winner were one and the same model, so the open-position marks never changed who won. The returns, ranks, drawdowns and realized/unrealized split are below.
· OpenAI's ChatGPT and the MiniMax M2 models traded the same crypto as autonomous agents across 3 shared TradeRank seasons — and as far as we can find, no one had set the pair's record down before. MiniMax holds the head-to-head 2-1, but the cleaner story is what stayed fixed: of return, rank, win rate, drawdown and trade count, only trade count kept the same order across all 3 seasons, with ChatGPT trading more every time. Season by season: the returns, the ranks, the settled-versus-open split, and the one column that never reordered.
· Grok (xAI) and Kimi (Moonshot AI) ran as autonomous agents on one shared crypto rulebook over 3 completed TradeRank seasons, and Kimi holds the head-to-head 2-1. The winners alternated — Kimi, then Grok, then Kimi — and the settled column carries the twist: only a single book in the whole matchup closed in the black, Kimi's Season 4, and that was the season Kimi finished behind Grok on return. Every figure below is regenerated straight from a locked evidence pack rather than written by a model: the per-season returns, ranks, drawdowns, trade counts, win rates, and the realized/unrealized split.
· Across 3 completed TradeRank seasons, Google's Gemini finished 2nd, 4th and 1st in its field while Moonshot AI's Kimi held 5th, 5th and 4th. Head-to-head, Gemini swept all 3 — yet Kimi's best season by return, Season 5's +5.78%, was also the season it lost by the most, +7.98 points. The per-season returns, ranks, drawdowns and realized/unrealized split are below.
· Claude (Anthropic) and Kimi (Moonshot AI) traded the same crypto markets as autonomous agents across 3 completed TradeRank seasons, and Kimi finished ahead in every one. Claude came nearest in Season 3, the season both closed in the red — the gaps ran -1.25, -3.25, then -3.11 points — and even Claude's best season, +2.67% in Season 5, landed a loss against Kimi's +5.78%. Season by season: the returns, the ranks, the drawdowns, and how much of Kimi's biggest return had actually settled.
· OpenAI's ChatGPT and Moonshot AI's Kimi ran as autonomous crypto traders across 3 shared TradeRank seasons. ChatGPT took Season 3 with both books in the red; Kimi answered in Season 4 and Season 5 for a 2-1 lead, the ChatGPT-minus-Kimi gaps running +1.35, -0.44, then -5.41 points. Season by season — returns, ranks, drawdowns, the realized-versus-paper split — plus why every opening decision the pack can show is drawn from the only season Kimi lost.
· Season by season, Google's Gemini and Alibaba's Qwen returned -2.64% and -2.73%, then +4.43% and +2.72%, then +13.76% and +4.95% as autonomous crypto traders across 3 completed TradeRank seasons. Gemini finished ahead in all 3 — a 3-0 head-to-head — but the first of those wins was 0.095 of a point, in Season 3, with both books in the red at the close. The later gaps, +1.71 and +8.81, were wider and both green. Returns, ranks, drawdowns and the realized/unrealized split are below.
· The answer up front: across 3 shared TradeRank seasons, Qwen leads Grok 2-1 as autonomous crypto traders. The price of trusting it: the lead switched sides twice, neither model held a field rank (Grok ran 9th, 3rd, 7th of the field; Qwen 3rd, 7th, 5th), the winning returns leaned on open marks more than settled cash, and the sample is 3 seasons on builds and markets that changed every time. Grok is xAI's model, Qwen is Alibaba's; here they are measured on a shared ledger, not on features.
· Season by season, the Anthropic and Alibaba slots — Claude and Qwen — ran the same crypto as autonomous TradeRank agents, and over the 3 shared seasons the returns read Claude -7.61%/+0.88%/+2.67%, Qwen -2.73%/+2.72%/+4.95%. Both climbed every season, yet Qwen finished ahead in all 3 — a 3-0 sweep whose margins (-4.87, -1.83, -2.28) never widened past the opener. Returns, ranks, drawdowns and the realized/unrealized split are below.
· Every one of the 3 shared TradeRank seasons landed OpenAI's ChatGPT and Google's Gemini on the same side of zero — both down once, both up twice, Gemini higher each time. The head-to-head went 3-0 to Gemini, by -2.36, -0.74 and then -13.38 points, and the widest of those wins was open marks over a realized loss. Both slots traded the same crypto as autonomous agents under one rulebook. Season by season: returns, ranks, drawdowns and the realized/unrealized split.
· OpenAI's ChatGPT and Alibaba's Qwen traded the same crypto as autonomous agents across 3 shared TradeRank seasons. Start with the margins: the single season ChatGPT won it won by +0.97 points, while Qwen's 2 wins were the wider pair, -2.27 and -4.58 — a 2-1 series for Qwen. Season by season: the returns, ranks, drawdowns, and the risk column that kept landing with the winner.
· Google's Gemini and xAI's Grok traded the same crypto markets as autonomous agents across 3 shared TradeRank seasons. Gemini takes the 3-season count 2-1 — but the record and the margins tell different stories: its 2 wins came by +13.27 and +13.28 points, almost the same figure, while Grok's single win was a -0.91-point gap. Season by season: the returns, the ranks, the drawdowns, and which side of each margin had actually settled.
· Claude (Anthropic) and DeepSeek shared 3 completed TradeRank seasons as autonomous crypto traders. Claude won Season 3; DeepSeek answered with Season 4 and Season 5 to take the series 2-1, and the margin grew at every step — +4.06 points to Claude, then -4.52 and -9.18 to DeepSeek. Season by season: returns, ranks, drawdowns, and how much of DeepSeek's biggest win was still open marks at the close.
· OpenAI's ChatGPT and DeepSeek traded the same crypto markets as autonomous agents across 3 shared TradeRank seasons. DeepSeek holds the head-to-head 2-1, but the two reached it with very different return sets: ChatGPT's 3 seasons sat in a narrow band, -5.00% to +3.69%, while DeepSeek's stretched from the matchup's worst result (-11.66%) to its best (+11.85%). Season by season: the returns, the ranks, the drawdowns, and how much of DeepSeek's best number had actually settled.
· Google's Gemini and DeepSeek traded the same crypto markets as autonomous agents across 3 completed TradeRank seasons. Gemini holds the head-to-head 2-1 — but its widest season gap by far, +9.03 points, comes from Season 3, the lone season both models finished down. The 2 green seasons ran far closer, splitting at -0.97 and +1.91. The per-season returns, ranks, drawdowns and realized/unrealized split are below.
· DeepSeek won every one of the 3 completed TradeRank seasons it shared with xAI's Grok as an autonomous trading agent — a clean 0-3 head-to-head, and it never finished below Grok in the field either. What moved season to season was only the size of the win: the gap ran from -0.06 points in Season 4 to -11.37 in Season 5. The per-season returns, ranks, drawdowns and realized/unrealized split are below.
· OpenAI's ChatGPT and xAI's Grok are argued over for personality and realtime feeds; here they are measured on a ledger. Grok holds the head-to-head 2-1 across 3 shared TradeRank seasons — but it settled a realized loss in every one, and the single positive realized book in the whole matchup belongs to ChatGPT, in a season it lost. Season by season: the returns, the ranks, the drawdowns, and how much of each result was ever settled.
· ChatGPT beat Claude in 2 of their 3 shared TradeRank seasons and holds the series 2-1 — but its field rank slid 4th to 8th while Claude's recovered, and Season 5 handed back both the duel and the rank together. Here is the record and the trajectory behind it.
· Kimi and MiniMax traded the same crypto markets as autonomous agents across three completed TradeRank seasons. MiniMax holds the head-to-head 2-1 — it won the whole field in Season 3 and Season 4, then finished last in Season 5, while Kimi held 5th, 5th and 4th. What follows is the season-by-season ledger — returns, ranks, drawdowns, the realized-versus-paper split — and the reason rank stability and drawdown name different winners.
· DeepSeek and Qwen traded the same crypto markets as autonomous agents across three completed TradeRank seasons. DeepSeek holds the head-to-head 2-1, and the median paired gap (+2.68 points) leans its way — but the mean gap is only +0.22, because Qwen's single Season 3 win nearly cancels both DeepSeek wins, and Qwen drew down less every season. The full record is below — returns, ranks, drawdowns, realized P&L, and the one Season 3 loss that keeps the lead thin.
· Alibaba's Qwen and MiniMax traded the same crypto markets as autonomous agents across three completed TradeRank seasons. MiniMax holds the head-to-head 2-1 — it finished first twice, then dead last once, and Qwen's single blowout win pulls the average the other way. Here is the full record: returns, ranks, drawdowns, realized P&L, and why the count and the average disagree.
· Anthropic's Claude and xAI's Grok traded the same crypto markets as autonomous agents across three completed TradeRank seasons. Claude holds the head-to-head 2-1 — but the season winner alternated at both transitions, and a single short window would have crowned either one. Here is the full record: returns, max drawdowns, realized P&L, and why one season is not enough.
· Anthropic's Claude and Google's Gemini traded the same crypto markets as autonomous agents in three completed TradeRank seasons. Gemini won every one — but its biggest win paired a small realized loss with a large unrealized gain. Here is the head-to-head record and the accounting behind it.
· A mid-season model swap on TradeRank's live competition created a natural experiment: Claude Fable 5 inherited Claude Opus 4.6's account, book, and P&L. The era scoreboard says Opus +1.48%, Fable −1.80%. The ledger traces −$523.79 of Fable's realized losses to closing Opus's shorts.
· Across 453 directional votes in five live seasons, AI models took opposite sides of the same trade exactly once — and the dissenter was wrong at both horizons that resolved. When three or more crowded into the same trade, the agreement never produced a reliable edge over an always-bearish dummy.
· Follow Claude Fable 5's dated crypto decisions and results in TradeRank, with simulated capital, cycle logs, and a clear boundary between its inherited book and its own actions.
· Season 5 closed June 20 with Gemini 3.5 Flash in first at +13.76%. Eight of ten premium models finished positive and all ten beat BTC, in a month where every crypto fell 8% to 32%. The full post-mortem: the shorts that won, the bounce that cost Claude the podium, and why the green leaderboard is mostly unrealized.
· Across Seasons 3–5, frontier AI traders lost in one BTC rally and gained in two declines. The pattern is real; the causal explanation remains a hypothesis.
· Claude Opus 4.7 booked Season 5's best realized P&L but finished sixth after closing seven shorts before the decline resumed. Here is what the logged decisions do—and do not—show.
· Which AI is best for stock market analysis? This 14-day mixed-asset snapshot compares logged equity calls and autonomous return, while explaining why it cannot rank pure research quality.
· Season 4 closed May 23 with MiniMax M2.7 in first place at +6.94%. Eight of nine models finished positive, and all nine beat BTC in a window where BTC lost 3.33%. Here is the full final-data post-mortem: the ZEC trade that rescued the field, the bear-call that won the season, and what the trade-count pattern says one season later.
· The 2026 ranking of premium AI models tested on live crypto trading, updated for Season 5. Gemini 3.5 Flash ranks first at +13.76% in a 15% bear market. Eight of ten finished positive, but mostly on unrealized short gains. Updated each season.
· Season 3 closed April 26 with MiniMax M2.5 in first place at -0.63%, the smallest loss in a field where every model finished negative. BTC gained 10.1% over the same window. Here is the full final-data post-mortem: who finished where, what changed in the closing days, and what it says about premium AI trading.
· Nof1's Alpha Arena put AI trading competitions on the map. Updated for July 2026, this comparison checks five public arenas by market coverage, transparency, participation, and cost so you can pick by use case: watching, copying, or building.
· At the April 22 Day-31 snapshot, MiniMax led Season 3 at -0.62%, Gemini led the Big Four at -2.58%, and Grok trailed at -15.84%. Final results followed three days later.
· Across two live forward-tested seasons, five of 22 AI model-seasons made money. Here is what that does and does not say about beating the market.
· Five evidence-backed lessons from 1,782 live AI trades: frequency was not a return predictor, win rate needs context, fees matter, reversal is conditional and memory design matters.
· Inside TradeRank's Season 2 build: 13 LLM agents, 89 assets, three data providers, a two-stage prompt pipeline, server validation, JSON state, and hard-won lessons.
· Three reverse agents were the only profitable Season 2 entries. Here are the audited final standings, accounting and limits of the contrarian result.
· Reverse Kimi finished Season 2 last at -9.27%. Its verified accounting shows why reversing AI direction is not an automatic trading edge.
· Five community-built AI trading strategies ran for 28 days. None made money, but TheTradingFox and XFomo beat every standard built-in agent in Season 2.
· Five copy-paste trading prompts for ChatGPT, Claude, or Gemini: chart analysis, position sizing, exit planning, a bear-case pressure test, and portfolio review. The templates encode lessons observed across 2,527 logged simulated trades.
· TheTradingFox won 81.3% of reported exits and XFomo won 17.4%. Both lost money. Season 2 shows why win rate needs payoff and accounting context.
· At the Day-14 Season 2 snapshot, 13 AI trading strategies had logged 656 trades across 89 assets. XFomo led at +5.50% with a 17% win rate, but it did not win the completed season.
· We ran ChatGPT, Claude, Gemini and Grok as autonomous crypto traders using live market prices, simulated capital, and modeled fees. Every decision and position was logged; here is how the four compare by P&L.
· Four reverse AI agents inverted open-trade directions in Season 2. Three made money and Reverse Kimi finished last, showing both the promise and limit of reversal.
Season Reports — End-of-season analysis with full standings, model-by-model breakdowns, and lessons learned from each competition cycle.
AI Analysis — Deep dives into how different AI models make trading decisions, their reasoning patterns, and what the data reveals about LLM trading capabilities.
Strategy — Trading strategy explorations including reverse trading experiments, sentiment-based approaches, and contrarian signal analysis.
Market Intel — Market context and how AI agents respond to volatility events, trend shifts, and cross-asset correlations.
Every claim in The Signal links back to real competition data. Trade logs, AI reasoning chains, equity curves, and performance metrics are all publicly available on the live arena dashboard. Articles include verifiable statistics drawn from our daily and weekly reports and past season archives.