Is Claude Sonnet 5.5 better than GPT-6.1 Sol?

On independent scores, yes; on cost per task, no. Artificial Analysis rates Claude Sonnet 5.5 (max effort) 56 on its Intelligence Index versus 52 for GPT-6.1 Sol (max), and Sonnet 5.5 leads Terminal-Bench 4.0 at 64% against 60% for Opus 5.5 and GPT-6 Astra. But Sonnet 5.5 at max uses ~193K output tokens per task — the most Artificial Analysis has ever measured — so it costs $7.60 per index task versus $0.72 for GPT-6.1 Sol, more than 10x, despite identical $2/$10 list prices. At high effort, Sonnet 5.5 sits narrowly behind GPT-6 Sol at about the same cost per task.

How much do Claude Sonnet 5.5 and GPT-6.1 Sol cost?

Both cost $2 per million input tokens and $10 per million output tokens as of September 30, 2026. Cache reads: Sonnet 5.5 $0.20, GPT-6.1 Sol $0.10 (cut from GPT-6 Sol's $0.20). Cache writes: Sonnet 5.5 $2.50 (5-minute), GPT-6.1 Sol $2.50. Both have 1M-token context and 128K max output. On a 30K-input / 5K-output task both are $0.11 at list; with 25K of the input cached, Sonnet 5.5 is about $0.065 and GPT-6.1 Sol about $0.0625. The real gap is tokens used per task, not price per token.

Which is faster, Claude Sonnet 5.5 or GPT-6.1 Sol?

Claude Sonnet 5.5 generates tokens faster: Artificial Analysis measured about 138 tokens per second for Sonnet 5.5 versus 76 for GPT-6 Sol and 66.8 for GPT-6.1 Sol at max effort. But because Sonnet 5.5 emits far more tokens per task at high effort, wall-clock time per task can still favor Sol. OpenAI's Ultrafast tier (up to 300 tokens/s at 6x price) is live for GPT-6 Astra and 'coming in the coming days' for GPT-6.1 Sol; Anthropic's fast mode is API-only for Opus, not Sonnet.

Which should I use for coding agents in 2026?

Claude Sonnet 5.5 if terminal-heavy agentic coding quality is the priority: 64% on Terminal-Bench 4.0 (Artificial Analysis) and 70.6% on Anthropic's own run, versus GPT-6 Astra's 60%. GPT-6.1 Sol if you run many agent turns and pay per token: OpenAI reports it matches GPT-6 Astra on DeepSWE v1.1 at one-fifth the cost, and its $0.10 cached input makes resent system prompts and tool lists nearly free. Run both on your own repo with real issues — there is no shared vendor benchmark between them.

Quick Answer

Claude Sonnet 5.5 vs GPT-6.1 Sol 2026: Same Price, Which?

Published: September 30, 2026

The short answer

Claude Sonnet 5.5 and GPT-6.1 Sol cost exactly the same — $2 per million input tokens, $10 per million output — and were released a day apart (September 28 and 29, 2026). Sonnet 5.5 is smarter on independent tests (Intelligence Index 56 vs 52); GPT-6.1 Sol is far cheaper per task ($0.72 vs $7.60 at max effort) because Sonnet 5.5 spends ~7x the tokens of GPT-6 Astra to get its score. Pick Sonnet 5.5 for terminal-heavy coding where quality per task matters; pick GPT-6.1 Sol for high-volume agents where you pay per token and reuse context. Facts verified September 30, 2026.

Side by side

	Claude Sonnet 5.5	GPT-6.1 Sol
Released	September 28, 2026	September 29, 2026 (DevDay)
Model id	`claude-sonnet-5-5`	`gpt-6.1-sol`
Input / output	$2 / $10	$2 / $10
Cache read	$0.20	$0.10
Cache write	$2.50 (5-min)	$2.50
Context / max output	1M / 128K	1.05M / 128K
Knowledge cutoff	June 2026	Not stated in launch post (GPT-6 Sol: April 20, 2026)
Effort levels	5 (low → max); API default high	none → max; default medium
AA Intelligence Index (max)	56 (#2)	52
AA cost per index task (max)	$7.60	$0.72
AA output tokens per task (max)	~193K (highest ever measured)	Not published; GPT-6 Astra ~1/7 of Sonnet’s
AA output speed	~138 tok/s	~67 tok/s
Terminal-Bench 4.0 (AA)	64%	Not on AA leaderboard yet; GPT-6 Astra 60%
Speed tier	Fast mode is Opus-only	Ultrafast (6x price) “coming days”
Clouds	AWS, GCP, Azure day one	OpenAI API; Bedrock via Managed Agents
Safety notes	First Sonnet with cyber safeguards; falls back to Sonnet 5 (~0.1% of tasks)	System card addendum; “no attempts to bypass an automated safety reviewer”

Scores are Artificial Analysis Intelligence Index v4.3.2 as read September 29–30, 2026. Sonnet 5.5 was evaluated on a pre-release build with a structured-output bug that Anthropic says is fixed in the public release; AA plans re-runs.

Price is identical; cost is not

Both vendors list $2/$10. Three things separate real bills:

Tokens per task. At max effort Sonnet 5.5 emits ~193K output tokens per Intelligence Index task — around 60% more than Opus 5.5 or Sonnet 5 at max, and ~7x GPT-6 Astra. That is why AA puts it “off the Intelligence vs. Cost per Task Pareto frontier.” At high effort the picture changes: AA says Sonnet 5.5 sits “very narrowly behind GPT-6 Sol on Intelligence at effectively the same cost per task.” So the effort knob matters more than the vendor.
Cache reads. GPT-6.1 Sol’s $0.10 is half Sonnet’s $0.20. For an agent resending a 25K-token system prompt and tool list every turn, that is $0.0025 vs $0.005 per turn — real at thousands of turns a day, invisible below that.
Speed tiers. OpenAI sells 2x Fast mode and 6x Ultrafast; Anthropic’s fast mode ($8/$40) is Opus 5.5 only. If you need a human-in-the-loop lane above 100 tokens/s, only Sonnet 5.5 gets there at list price today (~138 tok/s).

Reference task (30K in / 5K out) at list: $0.11 for both. Cost is decided by how many tokens the model chooses to think with, so benchmark on your workload at the effort level you will actually ship.

Where each wins

Claude Sonnet 5.5

Agentic terminal work: 64% Terminal-Bench 4.0 on AA (Sonnet 5 was 14%), above Opus 5.5 and GPT-6 Astra at 60%; Anthropic’s own run reports 70.6%.
Knowledge work parity with Opus 5.5: AA-Briefcase 1811 vs 1822 Elo, GDPval-AA 1844 vs 1846, AutomationBench-AA 71% vs 70%.
Lower hallucination rate than Opus 5.5 (47% vs 59%) though lower factual accuracy (54% vs 66%).
Anthropic says 30%+ faster output and up to 30% cheaper per task than Sonnet 5 through fewer tool calls — compare against AA’s higher measured token use before trusting either.

GPT-6.1 Sol

Cost per index task 10.5x lower at max effort; AA measured $0.72 vs $1.05 for GPT-6 Sol and $3.26 for Astra.
OpenAI-reported: matches Astra on DeepSWE v1.1 at one-fifth the cost; +2.2 pts over Opus 5.5 on AutomationBench at medium effort at a third of the cost; within 2.1 pts of Astra on OSWorld 2.0 at one-seventh the cost; factual-error rate at low effort down from 11.4% to 7.7% vs GPT-6 Sol.
Cheapest cache in the tier; Decisions API and Agents API integration on the same platform.
Caveat: GPT-6 Sol was replaced after seven days. Budget for model churn.

Migration notes

From Sonnet 5 → 5.5: thinking.disabled now returns 400 (use between_tools or low/medium/high), tool_choice: any|tool returns 400, cache minimum is 512 tokens, computer use moves to computer_toolset_20260801. Guide: how to migrate to Claude Sonnet 5.5.
From GPT-6 Sol → 6.1 Sol: same rate card, swap the model id; prompts over 272K tokens still reprice (2x input, 1.5x output for the whole request).
Both: pin the effort level in code. The difference between “medium” and “max” is bigger than the difference between the two vendors.

Verdict

Same list price, different bets. Sonnet 5.5 buys the highest score money can get at $2/$10 and pays for it in tokens; GPT-6.1 Sol buys 90% of the score at a tenth of the task cost and a cheaper cache. For a coding agent a developer watches, Sonnet 5.5. For a fleet of unattended agents billed by the token, GPT-6.1 Sol. Related: Sonnet 5.5 vs GPT-6 Sol vs Grok 4.7 and Sonnet 5.5 vs Opus 5.5.

Last verified: September 30, 2026. Prices from vendor pricing pages; scores from Artificial Analysis.