OpenAI, Google, Anthropic: Who Actually Won This Week in AI?

GPT-4o, Gemini 1.5 Pro, and Claude 3.5 Sonnet all just dropped upgrades—but only one system earned the crown. Developers, here’s what matters.

Last week, OpenAI’s GPT-4o scored a jaw-dropping 87.2% on the MMLU benchmark, beating Google’s Gemini 1.5 Pro (83.7%) and Anthropic’s brand-new Claude 3.5 Sonnet (81.1%). And it’s not just numbers: OpenAI slashed API prices by 50%. The result? There’s a new leader in language intelligence and cost.

What Just Shipped

• OpenAI GPT-4o: New version ships with 2x lower latency, 50% cheaper pricing. MMLU: 87.2%. Vision capabilities improved—now parses charts and diagrams reliably.
• Google Gemini 1.5 Pro: Public preview expands context window to 1M tokens, but latency remains high and pricing unchanged. MMLU: 83.7%.
• Anthropic Claude 3.5 Sonnet: Launched with improved coding and reasoning, but struggles on visual tasks. MMLU: 81.1%.
• xAI Grok-1.5: Still trailing: MMLU 73.2%, but rates high for uncensored answers and developer flexibility.

OpenAI’s price cut alone is a shot across the bow. GPT-4o now starts at $5 per million tokens for input and $15 per million tokens for output, half the previous rates. Google and Anthropic, for now, haven’t matched.

What’s Coming Next

• OpenAI: Voice mode for GPT-4o, agent actions (Project Glasswing), and multi-modal context expansion. Watch for OAuth-style authentication for AI agents—finally, a real answer to prompt injection and agent security.
• Google: Gemini 1.5 Pro’s 2M token context window is live in select APIs. Better vision and code understanding promised, but still vaporware for most users.
• Anthropic: Claude 3.5 Opus (the rumored flagship) expected within 3 months. Early demos suggest major leap in reasoning, but benchmarks pending.
• Microsoft: Copilot+ PC rollout brings local inference and “Recall” memory. Azure OpenAI gets GPT-4o soon—enterprise users, get ready.
• xAI: Grok-2 teased for Q3, but real-world performance remains speculative.

How This Changes Things

Let’s be blunt: OpenAI just made every developer’s workflow faster and cheaper. Before, GPT-4 was great but expensive. Now, GPT-4o is better, cheaper, and more flexible. Visual tasks? No contest—Google’s Gemini can’t reliably read hand-drawn charts, Claude still flubs diagrams, but GPT-4o nails it.

On coding, Claude 3.5 Sonnet gives solid explanations, but GPT-4o still generates cleaner, runnable code. Google’s Gemini 1.5 Pro shines in long context tasks (think legal or research docs), but unless you actually need million-token windows, GPT-4o wins on price and speed.

The Winner — With Evidence

GPT-4o: 87.2% (MMLU), $5/$15 per million tokens, near-instant responses.
Gemini 1.5 Pro: 83.7%, $10/$30, slow.
Claude 3.5 Sonnet: 81.1%, $8/$24, strong on code, weak on vision.
Grok-1.5: 73.2%, $5/$15, fast, but not reliable.

For 99% of real-world tasks, GPT-4o wins. The evidence is clear.

Actionable: Use GPT-4o’s New API Today

Want to try the new GPT-4o API? Here’s a quick-start for Python:

import openai
openai.api_key = 'YOUR_API_KEY'
response = openai.ChatCompletion.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Summarize the latest AI model upgrades."}]
)
print(response.choices[0].message['content'])

Swap in your API key. Enjoy faster, cheaper output. Done.

What Else Is Happening?

Microsoft Copilot+ PC is bringing local AI inference to laptops—no more cloud round-trips for basic tasks. xAI’s Grok is quietly growing on X, but still lags on benchmarks and enterprise adoption. Anthropic is hinting at agentic capabilities, but nothing concrete yet. The arms race isn’t slowing down, but this week, OpenAI took the lead and forced everyone else to react.

The Verdict

Ignore the hype. GPT-4o is now the default for anyone who needs fast, accurate results at scale. If you’re still betting on Google or Anthropic, check the benchmarks and your invoice. The numbers don’t lie.