GPT-6 Sol vs Claude Sonnet 5.5: Coding Agents & API Economics

Same Token Price, Different Cost per Task

OpenAI's GPT-6 Sol and Anthropic's Claude Sonnet 5.5 (released September 28, 2026) are the mid-tier models of their vendors' current lineups, and they list at the same $2 per million input tokens and $10 per million output tokens. Cached input costs $0.20 per million on GPT-6 Sol and $0.10 on Sonnet 5.5, after Anthropic halved Sonnet 5.5's cache reads on October 7, 2026. What you pay per finished task is a different matter. It depends on how many tokens each model spends to finish a task, on the reasoning effort you choose, and on how long your prompts are.

This guide keeps two kinds of evidence apart. Prices, limits and billing rules come from OpenAI's and Anthropic's own documentation. Benchmark results are labelled with who measured them: each vendor's own launch material, or the independent evaluator Artificial Analysis, which ran both models through the same test suite. The verdict at the end is ToolColumn's editorial judgment based on that evidence.

Official Prices and Limits

OpenAI released GPT-6.1 Sol on 2026-09-29 as an upgrade to GPT-6 Sol. It keeps the $2/$10 list price, the $2.50 cache write and the long-prompt multipliers, but bills cached input at $0.10 per million, half of GPT-6 Sol and the same as Sonnet 5.5 since Anthropic's October 7, 2026 cut. OpenAI reports gains over GPT-6 Sol, including 4.8 points on AutomationBench at medium effort and 6.4 points above GPT-6 Sol's best DeepSWE v1.1 result, but the independent measurements below cover GPT-6 Sol only. Retest GPT-6.1 Sol before applying them to it.

Two rules decide most bills before any benchmark does. First, long prompts: GPT-6 Sol reprices any request with more than 272K input tokens, while Sonnet 5.5 keeps standard rates across its 1M window. A single uncached request with 400,000 input tokens and 10,000 output tokens costs about $1.75 on Sol ($4 and $15 per million after the long-prompt multipliers) and about $0.90 on Sonnet 5.5. Second, tokenizers: Anthropic says Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text than its previous tokenizer, and OpenAI counts tokens with its own tokenizer. Identical per-token prices therefore do not guarantee identical bills for the same text.

Independent Head-to-Head: Artificial Analysis

Artificial Analysis is the only source in this guide that measured both models with the same method. Its comparison page sets Claude Sonnet 5.5 at max effort (adaptive reasoning, default fallback) against GPT-6 Sol at max effort. These are third-party measurements, not vendor figures, and they describe max effort only.

At max effort Sonnet 5.5 scores higher on the index and on every benchmark in this table, and it streams output faster, but it writes about six times as many output tokens per task. That makes it about seven times as expensive per task, and it takes longer to finish each task despite the faster stream. In its Sonnet 5.5 launch analysis, Artificial Analysis says Sonnet 5.5 at max costs about 50% more per task than Sonnet 5 and sits off its Intelligence versus Cost per Task Pareto frontier.

Max effort is not how most people run Sonnet 5.5, so effort matters more than the headline gap. Artificial Analysis also lists every Sonnet 5.5 effort level and every GPT-6 Sol effort level:

At each effort level Sonnet 5.5 scores higher and costs more per task. Matched by score instead, GPT-6 Sol is as cheap or cheaper: Sol at high (43, $0.38) outscores Sonnet 5.5 at medium (41, $0.59), and Sol at max (48, $1.05) edges Sonnet 5.5 at high (47, $1.08) for about the same cost. Only Sonnet 5.5 at xhigh or max scores above Sol's best result on this index.

Coding: Vendor Numbers Conflict

Neither vendor's table settles coding. In Anthropic's Sonnet 5.5 launch table, FrontierCode 1.1 (Main), which checks whether a code change could merge without human edits, has Sonnet 5.5 at 52.1% at Xhigh effort and 46.2% at Max, against 49.3% for GPT-6 Sol. Anthropic's footnote explains that at Max effort Sonnet 5.5 more often ran Claude Code's code-review skill across many subagents, which in some cases led to a timeout or to extra edits beyond the task's scope, and the benchmark penalizes those. Anthropic also says Sonnet 5.5 at High effort, the Claude Platform default, matches GPT-6 Sol's best FrontierCode score for about a fifth of the cost per task. OpenAI, for its part, says GPT-6 Sol matches Claude Fable 5.1 at xhigh on FrontierCode at much lower cost.

Anthropic reports Sonnet 5.5 at 70.6% on Terminal-Bench 4.0 and lists no GPT-6 Sol figure there. Artificial Analysis's own run measured 64% for Sonnet 5.5 and 44% for GPT-6 Sol at max effort. Anthropic also cautions that GPT-6 Sol's third-party knowledge-work scores may predate an OpenAI fix to Sol's image understanding; Artificial Analysis does not expect major changes. Treat all of these as signals to test, not a ranking: run both models on your own repository at the effort level you would actually ship.

Where Each Model Runs

GPT-6 Sol is available in the OpenAI API, and OpenAI made it and GPT-6 Luna available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Since 2026-10-07 ChatGPT Chat itself runs GPT-6 Sol on Plus and higher plans and GPT-6 Luna on Free and Go, so Free and Go users get Luna, not Sol. Sonnet 5.5 is available on the Claude API and is what the sonnet alias selects in Claude Code on the Anthropic API (Claude Code 2.1.284 or later). Claude Code is not part of the Claude Free plan. Claude's pricing page lists Sonnet for Free users without naming a version, so do not assume Free includes Sonnet 5.5.

Verdict: Which to Choose

This verdict is ToolColumn's editorial judgment, drawn from the official prices and billing rules and from Artificial Analysis's measurements above.

Pick GPT-6 Sol when cost per task matters, or when you already build on OpenAI and Codex. Per-token prices are identical, but in Artificial Analysis's runs Sol reached each score up to its ceiling for the same or less money per task, and at max effort it cost about a seventh as much per task as Sonnet 5.5.

Pick Claude Sonnet 5.5 when you need the stronger answer, when single prompts exceed 272K input tokens, or when your team works in Claude Code. Only Sonnet 5.5 at xhigh or max scored above Sol's best result, it streams output faster, and it bills long prompts at standard rates, which makes it cheaper by rule above 272K input tokens.

Whichever you run, set the effort level on purpose. On Artificial Analysis's index, Sonnet 5.5's cost per task ranges from $0.41 at low effort to $7.60 at max, and Sol's from $0.13 to $1.05. Effort moves the bill far more than the list price, which is the same for both.