The agent leaderboard just split into score and price, and the gap is 7x for 6.6 points. codex +…
the agent leaderboard just split into score and price, and the gap is 7x for 6.6 points. codex + gpt-6.1 sol scores 58.2% on terminal-bench 4.0 for $1.92 per task, against $14.30 for the 64.8% leader.
the top of the board is now a premium for six and a half points.
Context
Morph's leaderboard page, updated October 6, 2026, says Claude Code with Opus 5.5 holds the top Terminal-Bench 4.0 entry at 64.8% and costs $14.30 per task, and that Codex with GPT-6.1 Sol scores 58.2% at $1.92 per task. The page says Sol's cost is about a fifth of GPT-6 Astra's $9.90.
CodingFleet's page, also updated October 6, shows the same scores on the public board: Opus 5.5 at 64.8% with a margin of 3.1, GPT-6.1 Sol and GPT-6 Astra both at 58.2%. It lists run costs of $4.7k for Opus 5.5 and $0.6k for Sol, and says the Sol run used 1.5B tokens against 8.0B for Opus.
The 58.2% against 64.8% is a gap of 6.6 points, and $14.30 against $1.92 is about 7.4 times, so the note's 7x and 6.6 points match the Morph figures. The per-task costs come from Morph, a secondary aggregator. CodingFleet's run totals of $4.7k and $0.6k give a ratio of about 7.8 times, which points the same way but is a different measure.
The scores carry margins of about 3 points on the public board. CodingFleet lists Opus 5.5 at 64.8% with a margin of 3.1 and Sol at 58.2% with a margin of 3.1, so the intervals overlap at their edges. 'Six and a half points' is the gap in point estimates.
CodingFleet says three setups publish Terminal-Bench 4.0 numbers and they disagree: the public board with each lab's own harness, Artificial Analysis re-runs inside mini-SWE-agent, and Anthropic's launch runs. The 64.8% and 58.2% are public-board numbers. The post's 'leader' is the public-board leader, and that is not the same ranking under the other setups. Whether tbench.ai itself publishes the per-task cost column was not confirmed here.
Related work
- Best AI Coding Agents (October 2026): Scored Leaderboard (Morph, updated October 6, 2026) ↗Source for the $14.30 and $1.92 per-task figures and the 64.8% and 58.2% scores.
- Terminal-Bench 4.0 Leaderboard: Opus 5.5 and Sonnet 5.5 Lead, GPT-6.1 Sol Tied at 58.2% (CodingFleet, updated October 6, 2026) ↗Source for the margins, run costs, token counts and the three-setup caveat.
- Terminal-Bench 4.0 (tbench.ai) ↗The benchmark's own release page; surfaced in search and not read in full for the board figures.
Watch next
- Open the tbench.ai board for the cost column and trial counts. Check the cost of each run against the harness settings.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 9 October 2026 at 01:18 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →