gpt-6-astra vs gpt-6-sol
Parameters, independent leaderboard coverage, real-task reports and open-weight availability are shown side by side; missing coverage is not a zero.
Parameters
| Comparison item | gpt-6-astra | gpt-6-sol |
|---|---|---|
| Model Type | LLM | LLM |
| Context window | 1,050,000 tokens | 1,050,000 tokens |
| Maximum output | 128,000 tokens | 128,000 tokens |
| Input/output modalities | text, image → text | text, image → text |
| Endpoint | openai · openai-response | openai · openai-response |
| Open weights | Unavailable | Unavailable |
| License | Proprietary hosted model | Proprietary hosted model |
Leaderboard coverage
Every independent board covering either model is retained; an em dash means that board has not published a figure for that model.
A row marked as retrieved on different dates holds two readings taken on different days; a board re-fits or changes version as a whole between readings, so those two figures are not directly comparable.
| Evaluation | gpt-6-astra | gpt-6-sol |
|---|---|---|
| LMArena Text Arena — Arena score (Elo) | 1477 ±7 | 1457 ±7 |
| Artificial Analysis Intelligence Index — Intelligence Index Retrieved on different dates | 53 (retrieved 2026-09-21) | 48 (retrieved 2026-09-22) |
| Epoch Capabilities Index (ECI) — General ECI | 167 (163 - 171) | — |
| Vals Index — Accuracy (GDP-weighted finance, coding, legal and tax) | 65.02 ±1.10 | — |
| LMArena Text Arena · Chinese — Arena score (Elo) | 1510 ±22 | 1477 ±23 |
| LMArena Text Arena · Coding — Arena score (Elo) | 1543 ±13 | 1520 ±13 |
| Artificial Analysis · Output Speed — Median output tokens/s | 58 | 104 |
| Artificial Analysis · GDPval-AA v2.1 — Agentic Real-World Work Tasks, (Elo-500)/2000 (%) Retrieved on different dates | 52 (retrieved 2026-09-21) | 49 (retrieved 2026-09-22) |
| Artificial Analysis · Terminal-Bench v4.0 — Agentic Coding & Terminal Use (%) Retrieved on different dates | 59 (retrieved 2026-09-21) | 44 (retrieved 2026-09-22) |
| Artificial Analysis · MLCR-AA — Medical Long Context Reasoning (%) | 35.0 | — |
Real-task reports
gpt-6-astra
- Cortex AI: GPT-6 Astra, Gemini 3.8 Flash and Fable 5.1 on two real robot tasks — Load a sandwich into a microwave and close the door; attach a 5 mm tip to a pipette. Finding: GPT-6 Astra scored highest on both tasks, 80% on the microwave and 70% on the pipette, against 13.3% and 30% for Gemini 3.8 Flash and 6.7% and 20% for Fable 5.1; the lab credits it with retrying until the pipette tip docked and with checking that the microwave door latched. Limitation: The post does not say how many attempts each model had; scores are partial-credit judgements by the lab's evaluator, and the authors attribute the gap mainly to weaker depth perception in the lower-scoring models.
- GPT-6 Astra, Sol and Luna on one SVG animation prompt — Generate an SVG animation of a pelican riding a bicycle, shown in H5. Finding: GPT-6 Astra took 6 min 3 s and used 16% of the five-hour quota, against 4 min 42 s and 2% for GPT-6 Sol; the author recommends Sol when quota matters. Limitation: One prompt; the quota shares are a subscription meter, not API cost, and the post scores no output quality for Astra.
- Switching a coding project from GPT-6 Astra to DeepSeek V4.1 Flash — The author's own coding project, first on GPT-6 Astra and then on DeepSeek V4.1 Flash. Finding: About a dozen GPT-6 Astra turns cost the author $130 in a day without meeting the requirements; they switched the project back to DeepSeek V4.1 Flash. Limitation: One uncontrolled self-report; the post does not describe what the Astra turns were asked to do.
- Four models build the Eiffel Tower in Three.js — Build the Eiffel Tower in Three.js. Finding: GPT-6 Astra cost $7.45 and took 8 minutes; the author preferred Claude Opus 5.5's result ($8.95, 10 minutes) for this task. Limitation: One task judged by eye by the author; the post does not say how the models were run or how the costs were counted.
- Four models build a lava lamp from one prompt — Build a lava lamp from a single prompt. Finding: GPT-6 Astra cost $0.63 and took 5 min 4 s, between Claude Opus 5.5 ($1.27, 9 min 35 s) and GPT-6 Sol ($0.08, 50 s). Limitation: One prompt with the quality verdict left mostly to readers; the account is BridgeMind's own benchmark and does not say how the models were accessed.
- GPT-6 Sol and GPT-6 Astra on a robot drawing benchmark — The author's robot-drawing-on-the-board benchmark. Finding: GPT-6 Astra used about 22k output tokens to Sol's 25k, but Sol's controller drew twice as fast and cost almost five times less to run, though the author had expected stronger spatial reasoning from Sol. Limitation: One run of a benchmark the post describes only by name, with no score.
- AI Coding Daily LLM Coding Leaderboard — Implement and harden production-shaped projects in several stacks, including a Laravel app, a Go service and a React/TypeScript app, from the same prompts. Finding: GPT-6 Astra at medium reasoning scored 56.03 ($0.69 and 3 min 8 s per run in Codex CLI), third of 47 rows behind both Claude Opus 5.5 settings. Limitation: One author's benchmark whose code-quality half is graded by a judge model; rows run in different harnesses, the page is updated as models are added (read on 2026-09-24, 47 rows), and the author sells premium tutorials and advertising.
- kzhu's six-question LLM test leaderboard — Six questions scored out of 10: constrained sentence writing, two reasoning puzzles (replaced by pelican and compound-bow SVG animations in version 2.2), a browser OS with a space shooter, a 3D racing game and a Trello-style board. Finding: GPT-6 Astra scored 58/60 on version 2.1, ahead of Fable 5.1 (54), Gemini 3.8 Flash (49) and Qwen3.8 Max 0902 (47) on the same version: all reasoning answers right and a near-flawless browser OS, with only a collision glitch in the racing game. Limitation: One author's scores on six questions; rows use different harnesses and test versions (2.1 or 2.2), and a few rows' table total differs by one point from the author's note. This row does not name a harness.
gpt-6-sol
- Claude Opus 5.5 and GPT-6 Sol on one interactive website prompt — Build an interactive website about imaginary planets from a single prompt. Finding: GPT-6 Sol took 10 minutes and built three planets, using 1% of the $200 plan's usage; Claude Opus 5.5 took 26 minutes for a fuller, explorable solar system with more detail, using 16% of the $20 plan's usage. The author found the planets surprisingly similar: Sol faster, Opus bigger and more detailed. Limitation: One prompt judged by the author; the usage shares are meters of two different subscription plans, not API costs.
- Four models build the Eiffel Tower in Three.js — Build the Eiffel Tower in Three.js. Finding: GPT-6 Sol was the cheapest and fastest of the four at $3.90 and 5 minutes, but the author found its result a bit disappointing; they preferred Claude Opus 5.5 ($8.95, 10 minutes) and rated Kimi K3 ($4.04, 6 minutes) strong for the price. Limitation: One task judged by eye by the author; the post does not say how the models were run or how the costs were counted.
- Claude Pro, Codex Plus and SuperGrok on the same work at maximum settings — One piece of work run on each $20 subscription; the task itself is shown only in the author's video. Finding: GPT-6 Sol on Codex Plus used 9% of the weekly limit (29 million tokens) over 69 minutes and scored 7/10, the highest of the three plans, against 5/10 for Opus 5.5 and 6/10 for Grok 4.7; the author still recommends Claude at $20 because it yields far more tokens, and puts Codex Plus at about $330 of API usage a month. Limitation: The post does not describe the task, the quality score is the author's own, and the numbers are subscription meters rather than API usage.
- GPT-6 Astra, Sol and Luna on one SVG animation prompt — Generate an SVG animation of a pelican riding a bicycle, shown in H5. Finding: GPT-6 Sol took 4 min 42 s and used 2% of the five-hour quota, against 6 min 3 s and 16% for GPT-6 Astra; the author puts Sol's quota use at an eighth of Astra's and recommends Sol when quota matters. Limitation: One prompt; the quota shares are a subscription meter, not API cost, and the post gives no quality score for Sol.
- Four models build a lava lamp from one prompt — Build a lava lamp from a single prompt. Finding: GPT-6 Sol cost $0.08 and took 50 seconds, against $1.27 and 9 min 35 s for Claude Opus 5.5 and $0.63 and 5 min 4 s for GPT-6 Astra; in replies the account agreed Opus had a rough showing on this test. Limitation: One prompt with the quality verdict left mostly to readers; the account is BridgeMind's own benchmark and does not say how the models were accessed.
- Claude Opus 5.5 and GPT-6 Sol on a black hole merger test — The account's black hole merger visualisation test. Finding: GPT-6 Sol cost $0.08 and took 1 min 33 s, an eighth of Claude Opus 5.5's cost and four times faster, but the account judged Opus 5.5 'a significant leap for design. It's not close.' Limitation: One test judged by the account; the task is shown only in the attached video.
- Claude Opus 5.5 and GPT-6 Sol on one motion-graphic prompt — Make a 20-second motion graphic explaining how a language model turns a prompt into an answer and why long answers cost more. Finding: GPT-6 Sol took 11.4 minutes and $1.09 against Claude Opus 5.5's 12.7 minutes and $2.86; the author called the gap 'not even close' and agreed with a reply that Sol's result was far worse. Limitation: One subjective prompt; the verdict comes from the author's post and replies rather than a score.
- GPT-6 Sol and GPT-6 Astra on a robot drawing benchmark — The author's robot-drawing-on-the-board benchmark. Finding: GPT-6 Sol used about 25k output tokens against Astra's 22k and its controller drew twice as fast, at almost a fifth of the running cost; the author had expected stronger spatial reasoning from it. Limitation: One run of a benchmark the post describes only by name, with no score.
- AI Coding Daily LLM Coding Leaderboard — Implement and harden production-shaped projects in several stacks, including a Laravel app, a Go service and a React/TypeScript app, from the same prompts. Finding: GPT-6 Sol scored 52.42 at high reasoning ($0.31 and 5 min 18 s per run in Codex CLI) and 50.07 at medium, 8th and 16th of 47 rows; Claude Opus 5.5 led with 57.83 and GPT-6 Astra at medium scored 56.03. Limitation: One author's benchmark whose code-quality half is graded by a judge model; rows run in different harnesses, the page is updated as models are added (read on 2026-09-24, 47 rows), and the author sells premium tutorials and advertising.
- kzhu's six-question LLM test leaderboard — Six questions scored out of 10: constrained sentence writing, two reasoning puzzles (replaced by pelican and compound-bow SVG animations in version 2.2), a browser OS with a space shooter, a 3D racing game and a Trello-style board. Finding: GPT-6 Sol scored 44/60 on version 2.2 in Pi: full marks on the constrained sentences, but SVG animations with reversed wheels and an arrow that broke physics (6 each), a space shooter that only moved sideways, and an English interface for a Chinese Trello brief; Claude Opus 5.5 scored 55 on the same version. Limitation: One author's scores on six questions; rows use different harnesses and test versions (2.1 or 2.2), and a few rows' table total differs by one point from the author's note.