Gemini vs Claude for coding: benchmarks, price and agents

For coding in October 2026, Claude is ahead of Gemini on most independent measures, and Gemini is far cheaper than Claude's Sonnet and Opus. On LMArena's WebDev leaderboard, where developers vote on real builds, Claude Opus 5.5 is first at 1814; the best Gemini is Gemini 4 Argon, ninth at 1678, and the newest Gemini you can use today, 3.8 Flash, is thirtieth. Even Google's own benchmark table puts Opus 5.5 ahead of Argon on Terminal-Bench 4.0, 66.4% to 57.4%. On price, Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output through 2026 and has a free API tier, against $2 and $10 for Claude Sonnet 5.5 and $4 and $20 for Opus 5.5.

Key takeaways

• Quality: Claude leads LMArena WebDev, the Artificial Analysis Intelligence Index and Terminal-Bench 4.0.
• Price: Gemini 3.8 Flash is about a third of Sonnet 5.5's price, and free on the Gemini API's free tier.
• Gemini 4 Argon, announced September 30, is limited to selected cyber defenders; paid API customers and Google AI Ultra come next, with no date.
• Context: both labs' current models take 1 million input tokens.
• Agents: Claude Code comes with Claude Pro at $20; Google's Antigravity comes with Google AI Pro at $19.99 and also offers Claude Sonnet and Opus 5.5 to paying subscribers.

Is Gemini or Claude better for coding?

Claude, on the independent rankings that cover both:

• LMArena WebDev, October 7, 2026 (LMArena): Opus 5.5 1814 (1st), Sonnet 5.5 1773 (3rd), Gemini 4 Argon 1678 (9th, preliminary), Gemini 3.7 Flash 1592 (29th), Gemini 3.8 Flash 1583 (30th).
• Artificial Analysis Intelligence Index (Artificial Analysis): Opus 5.5 57.6, Sonnet 5.5 56.0, Gemini 4 Argon 52.6, Gemini 3.8 Flash 40.9, Gemini 3.1 Pro 29.7.
• Terminal-Bench 4.0, on the same harness (Vals AI): Opus 5.5 65.15%, Sonnet 5.5 64.14%, Gemini 4 Argon 57.58%.

Google's own table has Argon ahead on two benchmarks, below. Our comparison with OpenAI's models is in Claude vs ChatGPT for coding, and the two Claude models in Sonnet 5.5 vs Opus 5.5.

What about Gemini 4 Argon?

Google announced Gemini 4 Argon on September 30, 2026. It's "rolling out to a set of trusted cyber defenders" through Google's Fairwind Program, and comes to paid API customers and Google AI Ultra subscribers next (Google). Google's launch price is $2 per million input tokens and $10 output, rising to $4 and $20 after an introductory period.

Google's own numbers, which it computed for Argon with its own harness (Google DeepMind):

• Argon ahead: DeepSWE v1.1, 77.9% to Opus 5.5's 74.2%; Vibe Code Bench, 91.9% to 90.3%.
• Opus 5.5 ahead: Terminal-Bench 4.0, 66.4% to 57.4%; FrontierSWE v2, 62.3% to 55.0%.

Until Argon is public, the Gemini you can code with is 3.8 Flash or 3.1 Pro.

How much do Gemini and Claude cost for coding?

API prices per million input and output tokens, October 2026 (Google, Anthropic):

• Gemini 3.8 Flash: $0.75 and $3.75 through December 31, 2026, then $1.50 and $7.50. Free on the free tier, where your data is "used to improve our products."
• Gemini 3.1 Pro (preview): $2 and $12 up to 200K tokens, $4 and $18 above. Not on the free tier.
• Claude Haiku 5.5: $0.10 and $0.50 up to 100K tokens, $0.50 and $2.50 above. Released October 7, 2026, it costs less per token than Gemini 3.8 Flash.
• Claude Sonnet 5.5: $2 and $10.
• Claude Opus 5.5: $4 and $20.

For heavy volume, such as reviewing every pull request or running many agents in parallel, Gemini Flash's price can matter more than its benchmark gap.

Claude Code vs Gemini CLI and Antigravity

• Gemini CLI stopped serving free and Google AI Pro and Ultra accounts on June 18, 2026; Google moved them to the Antigravity CLI, and Gemini CLI now needs a paid API key or a business license (Gemini CLI free tier).
• Gemini Code Assist in IDEs now serves paid Standard and Enterprise licenses only (Gemini Code Assist pricing).
• Antigravity, Google's agent IDE and CLI, runs Gemini 3.8, 3.7 and 3.6 Flash and Gemini 3.1 Pro on every plan, and Claude Sonnet and Opus 5.5 on Google AI Pro (not trials) and Ultra (Google). On AI Pro the quota refreshes every five hours up to a weekly limit.
• On Artificial Analysis's coding agents index, Claude Code with Sonnet 5.5 scores 68, the Antigravity CLI with Gemini 4 Argon 64 (not yet public), and Antigravity's SDK with Gemini 3.8 Flash 42.

Full comparison in Antigravity vs Claude Code.

Which do developers use?

In Stack Overflow's 2026 survey, 74.6% of developers who use AI at work use Anthropic's models and 54.5% Google's, out of 11,895 answers (Stack Overflow). Among coding agents, 65.5% had used Claude Code in the past year, 16.0% Google Antigravity and 14.6% Gemini Code Assist (Stack Overflow).

Should you use Gemini or Claude?

• Claude for the hardest code, long agent sessions and work you'll ship without rewriting. Sonnet 5.5 for most of it, Opus 5.5 for the hardest.
• Gemini 3.8 Flash for high-volume or budget work, prototypes and learning, especially on the free tier.
• Both: many teams use one to write and the other to review (Cross-model code review).

Comparing models as a team

The fastest way to settle Gemini vs Claude is your own codebase. Poly (usepoly.co) is a shared browser room where a team works with one AI agent, everyone sees each step live, any member can approve a change, and the room can switch between models from several labs mid-conversation. It runs Claude Code and Codex. Free to start. What is Poly?