GPT-5.6 Won the Headlines. The Money Bet on Anthropic.
On Polymarket's deepest-liquidity AI market â $2.27 million in real money on the table â traders give Anthropic a 94% chance of having the best AI model at the end of July 2026. Google gets 5%. OpenAI gets 1%. One percent. On the same day Sam Altman's victory-lap tweet about GPT-5.6 Sol being "the best model in the world right now" pulled 2.58 million views.
đ Read the full version with charts and embedded sources on ComputeLeap â
That divergence â between the loudest signal in AI (YouTube thumbnails, X engagement, launch-day fireworks) and the money signal (prediction markets, IPO discourse, enterprise contracts) â is the story nobody's writing. Everyone's covering the launch. Nobody's asking why the people with skin in the game aren't buying it.
The Scoreboard That Pays Out
Polymarket isn't a poll. It's a prediction market where traders put real capital behind their convictions and lose real money when they're wrong. The "Which company has best AI model end of July?" market has been running all year, and the numbers tell a story that looks nothing like your YouTube feed:
- Anthropic: 94% (down from 100% earlier in June, but still a coronation)
- Google: 5% (the math-model dark horse)
- OpenAI: 1% (the company that just launched GPT-5.6)
That 1% is not a rounding error. It's $2.27 million worth of collective conviction that GPT-5.6 â for all its benchmark claims and government-coordination drama â does not change the leaderboard. We covered this market when Anthropic was at 92%; it's only gotten more lopsided since.
âšī¸ Polymarket's AI model market is the deepest-liquidity prediction market in the AI category â $2.27M total with $237K in 24-hour volume on July 11 alone. These aren't retail gamblers; this is informed capital with weekly P&L statements.
The June-end market told the same story: Anthropic at 94.8%, with $16 million total traded across the question's lifetime. Claude Fable 5, Claude Opus 4.8, and their thinking variants have held the top four spots on composite intelligence indices since May. GPT-5.5 sits fifth. GPT-5.6 launched two days ago and hasn't moved the needle.
The Victory Lap That Fooled Nobody (With Money)
Sam Altman's GPT-5.6 announcement is a masterclass in tech CEO theater: "there are a lot of benchmarks that suggest 5.6 sol is the best model in the world right now, but the most reliable way to tell is that elon is obsessed with me again." 42,700 likes. 2.9K retweets. 2.58 million views. YouTube creators scrambled to publish takes within hours of the launch.
The launch was coordinated â CNBC reported that GPT-5.6 was gated behind a government safety review before going public on July 9, with Sol (frontier reasoning), Terra (balanced), and Luna (fast/cheap) as the tier names. The media narrative wrote itself: safety-conscious release, tiered pricing, benchmarks above 5.5.
But here's what didn't happen: the Polymarket odds didn't move. The day GPT-5.6 went live, Anthropic's share of the "best model" market held steady at 94%. The traders who had weeks of advance notice about the launch â it was the worst-kept secret in AI â had already priced in everything they expected. Their price: 1%.
The $3 Trillion Thesis
While GPT-5.6 was eating the timeline, the real money conversation was happening on the All-In Podcast. In Episode 278, with Brad Gerstner filling in for Friedberg, investor Gavin Baker said the quiet part loud: "I think Anthropic is worth $3 trillion today."
His framework is specific and falsifiable:
- End 2026 with over $100B in annual revenue
- Reach $200-300B revenue in 2028
- Hold 85% gross margins on inference
- Apply a 10x multiple
The numbers aren't fantasy. Anthropic's annualized revenue hit $47 billion in May â the fastest ramp in enterprise software history. From $87 million in January 2024 to $47 billion in 28 months. Salesforce took 20 years to reach $30B. AWS took 13. Anthropic did it before filing its S-1 at a $965B valuation.
Gerstner called it "the revenue ramp we've never seen in enterprise software." Chamath doubled down with the enterprise-moat thesis: production systems for large, regulated enterprises where "vibing isn't tolerated â these are the systems that run western society: banking, power, healthcare, insurance."
The subtext: enterprises don't switch AI providers because a new model scored 2 points higher on a benchmark. They switch when contracts expire, when security reviews complete, when compliance teams sign off. Anthropic's 80% enterprise revenue share isn't a benchmark â it's a moat.
Dwarkesh's Question: The Only One That Matters
The sharpest framing of the entire cycle came not from a VC but from podcast host Dwarkesh Patel, who wrote what amounts to the article everyone else is dancing around:
This is the entire game stated in one sentence. The three-lab equilibrium (Anthropic, OpenAI, Google) keeps prices low. The moment one lab pulls ahead durably â not for a launch week, but for a fiscal quarter â the winner gets monopoly-adjacent pricing. And the Polymarket odds suggest that moment may have already arrived.
The contrarian case for OpenAI? CNBC's model-routing piece argues pricing power is shifting from sellers to buyers regardless. If enterprises route easy tasks to cheap open-source models and only send hard problems to frontier labs, even the best model doesn't capture the whole market. But that argument cuts against all frontier labs equally â it doesn't explain why the market prices Anthropic 93 points above OpenAI.
What the Community Is Saying
The Hacker News thread "Anthropic surpasses OpenAI to become most valuable AI startup" (422 points, 472 comments) captured the developer zeitgeist perfectly. The top comments debate whether Claude's dominance is "marketing" or "genuinely superior agentic capabilities" â but notably, nobody disputes the valuation crossing.
Meanwhile, "Leaked OpenAI financials show $38.5B loss and compute burn" (221 points) revealed the other side of the ledger. HN commenters noted that OpenAI's $13B revenue against $7.5B cost of revenue makes inference appear profitable â but profitable inference doesn't help if you're losing the enterprise sales war.
â ī¸ Contrarian Corner: The prediction market might be wrong. GPT-5.6 launched 48 hours ago â markets are backward-looking by nature, pricing last month's arena results. If Sol genuinely outperforms on enterprise workloads over the next 2-3 weeks, the 94/1 split could narrow fast. The 6.8% monthly dip in Anthropic's share shows the market isn't completely static. And OpenAI's distribution moat (ChatGPT's hundreds of millions of users) doesn't show up on any benchmark â but it shows up in revenue.
The Signals Diverge: A Visual Summary
Here's what the two signal types are telling you about the same week:
| Signal Type | What It Says | Evidence |
|---|---|---|
| Hype signals (views, thumbnails, engagement) | GPT-5.6 is the story of the week | Sam's tweet: 2.58M views. 5+ YouTube videos in 24h. CNBC front page. |
| Money signals (prediction markets, valuations, enterprise contracts) | Anthropic owns the cycle | Polymarket: 94% vs 1%. $47B ARR. $965B valuation. $3T IPO thesis on All-In. |
This isn't the first time hype and money have diverged in tech â crypto taught us that lesson repeatedly. But in AI, the divergence has a specific mechanism: launch-day benchmarks move engagement; enterprise contracts move revenue. And revenue is what VCs price.
What This Means for You
đĄ If you're choosing an AI provider for production systems: The market is telling you that benchmark leaderboard position is transient but platform lock-in is durable. Today's "best model" title changes quarterly; your API integration, fine-tuning investment, and compliance certifications don't. Follow the enterprise money, not the X timeline.
If you're an investor or following the IPO: The $3T Anthropic thesis requires two things: (1) revenue continuing its vertical ramp past $100B/year, and (2) the three-lab equilibrium breaking in Anthropic's favor so pricing power kicks in. Polymarket says condition #2 is already met. The S-1 will tell us about condition #1.
The real bear case isn't GPT-5.6 â it's model routing plus open-source commoditization shrinking the total addressable market for premium inference. 247 Wall Street's analysis of IPO prediction markets shows Anthropic is also the favorite in the "which AI lab IPOs first" race at 78 cents â the market sees its corporate structure clearing regulatory hurdles faster.
If you're building content or narratives around AI: The engagement-to-truth ratio in AI coverage has never been worse. A 2.58M-view tweet and a $2.27M prediction market are telling opposite stories. One of them is wrong. Historically, the people with money on the line are right more often than the people optimizing for likes.
The prediction market could be wrong â it's been wrong before. But $2.27 million in liquidity is a more honest signal than 2.58 million views. Views are free. Bets cost money. And right now, the money is speaking clearly: GPT-5.6 won the week. Anthropic won the year.
Originally published at ComputeLeap
Top comments (0)