# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/49817404

# GPT\-6 Astra has gained the ability to drive a car

198 points · 169 comments

[Full discussion](<https://news.ycombinator.com/item?id=49817404>)

[Read original](<https://drivingbench.com/>)

Category: [Models & Products](<https://hacksnap.live/?category=models-products>)

## Skept-o-meter & Hotness

Skept\-o\-meter: High\. Estimated from 10 comments\.

22 comments for the summary\.

Peak rank: \#4

Time in Top 10: 24\.9 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

227 recorded rank observations from 2026\-09\-23T19:09:03\.485431\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 168 recorded Hacksnap ranks:

2026\-09\-30T18:03:34\.83179\+00:00: rank \#95

2026\-09\-30T19:02:02\.291882\+00:00: rank \#95

2026\-09\-30T20:02:49\.392147\+00:00: rank \#96

2026\-09\-30T21:03:42\.990832\+00:00: rank \#94

2026\-09\-30T22:02:11\.294387\+00:00: rank \#95

2026\-09\-30T23:01:52\.722953\+00:00: rank \#95

2026\-10\-01T08:02:24\.797064\+00:00: rank \#95

2026\-10\-01T09:01:35\.899477\+00:00: rank \#98

2026\-10\-01T10:01:12\.485692\+00:00: rank \#98

2026\-10\-01T11:00:47\.508128\+00:00: rank \#98

2026\-10\-01T12:00:42\.477917\+00:00: rank \#97

2026\-10\-01T13:00:51\.124852\+00:00: rank \#97

2026\-10\-01T14:00:59\.01153\+00:00: rank \#97

2026\-10\-01T15:00:51\.048047\+00:00: rank \#98

2026\-10\-01T16:00:53\.034654\+00:00: rank \#99

2026\-10\-01T16:36:37\.180071\+00:00: rank \#100

2026\-10\-01T17:00:23\.382263\+00:00: rank \#100

2026\-10\-01T18:01:44\.046011\+00:00: rank \#101

2026\-10\-01T19:01:47\.420738\+00:00: rank \#102

2026\-10\-01T20:01:39\.585083\+00:00: rank \#101

2026\-10\-01T21:01:45\.872085\+00:00: rank \#102

2026\-10\-01T22:01:37\.214899\+00:00: rank \#101

2026\-10\-01T23:04:11\.395288\+00:00: rank \#104

2026\-10\-02T08:02:26\.819307\+00:00: rank \#105

2026\-10\-02T09:02:51\.51649\+00:00: rank \#107

2026\-10\-02T10:02:12\.346099\+00:00: rank \#108

2026\-10\-02T11:02:22\.794376\+00:00: rank \#108

2026\-10\-02T12:01:14\.681578\+00:00: rank \#108

2026\-10\-02T13:02:29\.511591\+00:00: rank \#108

2026\-10\-02T14:01:40\.936738\+00:00: rank \#107

2026\-10\-02T15:02:30\.225476\+00:00: rank \#108

2026\-10\-02T16:02:47\.060695\+00:00: rank \#107

2026\-10\-02T17:02:31\.351493\+00:00: rank \#109

2026\-10\-02T18:01:22\.271899\+00:00: rank \#110

2026\-10\-02T19:01:23\.681417\+00:00: rank \#111

2026\-10\-02T20:01:33\.801055\+00:00: rank \#113

2026\-10\-02T21:02:55\.453316\+00:00: rank \#116

2026\-10\-02T22:01:50\.233139\+00:00: rank \#117

2026\-10\-02T23:01:20\.268516\+00:00: rank \#115

2026\-10\-03T08:01:30\.186188\+00:00: rank \#116

2026\-10\-03T09:01:57\.91467\+00:00: rank \#116

2026\-10\-03T10:00:55\.732001\+00:00: rank \#115

2026\-10\-03T11:01:22\.538596\+00:00: rank \#115

2026\-10\-03T12:01:08\.789384\+00:00: rank \#116

2026\-10\-03T13:00:35\.273559\+00:00: rank \#116

2026\-10\-03T14:00:48\.425606\+00:00: rank \#116

2026\-10\-03T15:01:04\.472963\+00:00: rank \#116

2026\-10\-03T16:01:56\.56175\+00:00: rank \#117

2026\-10\-03T17:01:36\.651036\+00:00: rank \#114

2026\-10\-03T18:01:58\.24207\+00:00: rank \#114

2026\-10\-03T19:01:19\.196589\+00:00: rank \#114

2026\-10\-03T20:00:57\.313759\+00:00: rank \#113

2026\-10\-03T21:01:04\.44439\+00:00: rank \#112

2026\-10\-03T21:20:01\.585706\+00:00: rank \#112

2026\-10\-03T22:00:15\.503126\+00:00: rank \#112

2026\-10\-03T23:00:55\.584832\+00:00: rank \#112

2026\-10\-04T08:01:00\.327459\+00:00: rank \#112

2026\-10\-04T09:00:59\.321023\+00:00: rank \#113

2026\-10\-04T10:00:42\.975336\+00:00: rank \#113

2026\-10\-04T11:01:03\.76577\+00:00: rank \#113

2026\-10\-04T12:01:45\.307985\+00:00: rank \#113

2026\-10\-04T13:00:49\.176267\+00:00: rank \#113

2026\-10\-04T14:00:56\.952543\+00:00: rank \#113

2026\-10\-04T15:00:54\.739637\+00:00: rank \#114

2026\-10\-04T16:00:51\.403054\+00:00: rank \#115

2026\-10\-04T17:00:43\.166713\+00:00: rank \#115

2026\-10\-04T18:00:46\.315535\+00:00: rank \#114

2026\-10\-04T19:00:54\.768841\+00:00: rank \#114

2026\-10\-04T20:00:51\.070442\+00:00: rank \#115

2026\-10\-04T21:00:28\.550532\+00:00: rank \#114

2026\-10\-04T22:01:36\.217651\+00:00: rank \#116

2026\-10\-04T23:01:09\.400767\+00:00: rank \#116

2026\-10\-05T08:01:51\.387003\+00:00: rank \#116

2026\-10\-05T09:01:40\.528475\+00:00: rank \#116

2026\-10\-05T10:00:56\.290772\+00:00: rank \#116

2026\-10\-05T11:01:05\.919518\+00:00: rank \#116

2026\-10\-05T12:00:56\.590881\+00:00: rank \#116

2026\-10\-05T13:01:05\.197156\+00:00: rank \#118

2026\-10\-05T14:01:04\.085718\+00:00: rank \#118

2026\-10\-05T15:00:50\.814571\+00:00: rank \#118

2026\-10\-05T16:02:09\.54139\+00:00: rank \#117

2026\-10\-05T17:00:35\.336076\+00:00: rank \#117

2026\-10\-05T18:00:20\.445491\+00:00: rank \#117

2026\-10\-05T19:02:43\.604975\+00:00: rank \#118

2026\-10\-05T20:01:09\.777459\+00:00: rank \#118

2026\-10\-05T21:01:47\.650671\+00:00: rank \#119

2026\-10\-05T22:02:59\.93099\+00:00: rank \#120

2026\-10\-06T08:02:11\.210914\+00:00: rank \#120

2026\-10\-06T08:02:14\.354697\+00:00: rank \#120

2026\-10\-06T09:01:21\.688765\+00:00: rank \#122

2026\-10\-06T10:00:58\.047773\+00:00: rank \#122

2026\-10\-06T11:00:41\.689929\+00:00: rank \#123

2026\-10\-06T12:00:41\.7705\+00:00: rank \#124

2026\-10\-06T13:01:24\.509519\+00:00: rank \#124

2026\-10\-06T14:00:26\.950817\+00:00: rank \#124

2026\-10\-06T15:00:49\.066414\+00:00: rank \#125

2026\-10\-06T16:00:58\.166747\+00:00: rank \#127

2026\-10\-06T17:01:17\.009511\+00:00: rank \#127

2026\-10\-06T18:01:43\.791429\+00:00: rank \#128

2026\-10\-06T19:00:46\.048609\+00:00: rank \#128

2026\-10\-06T20:01:14\.687294\+00:00: rank \#129

2026\-10\-06T21:02:13\.414382\+00:00: rank \#129

2026\-10\-06T22:00:40\.926271\+00:00: rank \#128

2026\-10\-06T23:03:26\.867445\+00:00: rank \#130

2026\-10\-07T08:01:41\.339804\+00:00: rank \#131

2026\-10\-07T09:02:38\.026986\+00:00: rank \#133

2026\-10\-07T10:01:35\.069531\+00:00: rank \#133

2026\-10\-07T11:02:07\.770084\+00:00: rank \#133

2026\-10\-07T12:01:14\.996697\+00:00: rank \#132

2026\-10\-07T13:00:47\.45912\+00:00: rank \#131

2026\-10\-07T14:01:18\.307739\+00:00: rank \#131

2026\-10\-07T15:01:57\.541362\+00:00: rank \#134

2026\-10\-07T16:01:28\.474022\+00:00: rank \#134

2026\-10\-07T17:01:15\.036908\+00:00: rank \#135

2026\-10\-07T18:00:53\.300079\+00:00: rank \#135

2026\-10\-07T19:01:49\.456657\+00:00: rank \#136

2026\-10\-07T20:01:59\.032487\+00:00: rank \#136

2026\-10\-07T21:05:36\.655392\+00:00: rank \#138

2026\-10\-07T22:02:27\.105241\+00:00: rank \#139

2026\-10\-07T23:01:17\.452894\+00:00: rank \#139

2026\-10\-08T08:03:55\.083808\+00:00: rank \#139

2026\-10\-08T09:03:58\.68898\+00:00: rank \#139

2026\-10\-08T10:04:05\.364235\+00:00: rank \#140

2026\-10\-08T11:02:04\.410307\+00:00: rank \#139

2026\-10\-08T12:02:36\.39618\+00:00: rank \#139

2026\-10\-08T13:03:09\.206401\+00:00: rank \#139

2026\-10\-08T14:03:30\.113384\+00:00: rank \#140

2026\-10\-08T15:02:54\.444758\+00:00: rank \#138

2026\-10\-08T16:03:13\.067807\+00:00: rank \#137

2026\-10\-08T17:03:47\.92465\+00:00: rank \#136

2026\-10\-08T18:01:50\.360431\+00:00: rank \#136

2026\-10\-08T19:02:55\.719274\+00:00: rank \#137

2026\-10\-08T20:01:36\.359022\+00:00: rank \#137

2026\-10\-08T21:03:34\.000719\+00:00: rank \#139

2026\-10\-08T22:01:13\.459732\+00:00: rank \#139

2026\-10\-08T23:01:25\.318392\+00:00: rank \#138

2026\-10\-09T08:01:53\.314021\+00:00: rank \#139

2026\-10\-09T09:02:59\.93426\+00:00: rank \#140

2026\-10\-09T10:02:22\.652527\+00:00: rank \#140

2026\-10\-09T11:02:04\.807282\+00:00: rank \#140

2026\-10\-09T12:02:16\.769295\+00:00: rank \#140

2026\-10\-09T13:02:30\.35034\+00:00: rank \#142

2026\-10\-09T14:01:34\.215134\+00:00: rank \#142

2026\-10\-09T15:01:27\.564392\+00:00: rank \#142

2026\-10\-09T16:02:06\.561056\+00:00: rank \#143

2026\-10\-09T17:03:06\.673648\+00:00: rank \#145

2026\-10\-09T18:01:02\.851052\+00:00: rank \#145

2026\-10\-09T19:01:11\.250291\+00:00: rank \#146

2026\-10\-09T20:01:32\.91907\+00:00: rank \#146

2026\-10\-09T21:02:08\.376144\+00:00: rank \#148

2026\-10\-09T22:01:19\.253075\+00:00: rank \#150

2026\-10\-09T23:01:31\.990912\+00:00: rank \#150

2026\-10\-10T08:01:00\.593376\+00:00: rank \#149

2026\-10\-10T09:01:45\.527975\+00:00: rank \#151

2026\-10\-10T10:00:53\.799353\+00:00: rank \#151

2026\-10\-10T11:00:35\.65851\+00:00: rank \#150

2026\-10\-10T12:01:11\.169647\+00:00: rank \#150

2026\-10\-10T13:00:54\.008377\+00:00: rank \#150

2026\-10\-10T14:01:43\.208766\+00:00: rank \#152

2026\-10\-10T15:01:24\.825188\+00:00: rank \#152

2026\-10\-10T16:01:10\.213328\+00:00: rank \#151

2026\-10\-10T17:00:48\.84189\+00:00: rank \#150

2026\-10\-10T18:00:57\.55132\+00:00: rank \#150

2026\-10\-10T19:00:52\.379463\+00:00: rank \#149

2026\-10\-10T20:00:21\.993912\+00:00: rank \#149

2026\-10\-10T21:01:10\.657738\+00:00: rank \#150

2026\-10\-10T22:00:50\.433569\+00:00: rank \#149

2026\-10\-10T23:01:00\.95027\+00:00: rank \#150

The benchmark is striking as a demonstration that a frontier LLM can complete a simple course, but the discussion is dominated by doubt that cloud\-latency, robustness, and interpretability problems make it a path to real\-world self\-driving\.

## The brief

DrivingBench is a leaderboard for frontier LLMs driving a real car around a course, allowing up to 3 attempts within one continuous chat\. GPT\-6 Astra (Codex, medium) is the only listed model to finish, reaching 100% on its second attempt in 5:22 at a cost of 246\.6M tokens / $7\.74, after a first attempt that reached 49% and DNF'd\. Claude Fable 5\.1 (Claude Code, medium) peaked at 45% (DNF), Grok 4\.6 (Cursor, medium) at 11%, and GPT\-5\.6 Sol (Codex, medium) at 6%\. Progress is measured as how far along the course centerline an attempt got while staying within 4 m of it, with a collision preserving progress reached before it; other metrics include GPS\-integrated distance, finish time, accepted set\_motion/stop\_now commands, and token/cost totals\.

- GPT\-6 Astra (Codex, medium) is the only model shown finishing the course, hitting 100% on attempt 2 in 5:22, using 246\.6M tokens at $7\.74\.
- Astra's first attempt in the same chat reached 49% and DNF'd, costing 81\.2M tokens / $2\.01\.
- Claude Fable 5\.1 (Claude Code, medium) peaked at 45% (DNF) across three attempts, with earlier attempts at 9% and 10%\.
- Grok 4\.6 (Cursor, medium) peaked at 11% and GPT\-5\.6 Sol (Codex, medium) at 6%, both DNF on all attempts\.
- Progress is defined as how far along the course centerline an attempt got while staying within 4 m of it, as a share of centerline length to the finish zone, with collisions preserving prior progress\.
- The benchmark permits up to 3 attempts in one continuous chat and reports GPS\-integrated distance, finish time, accepted set\_motion/stop\_now commands, and token/cost totals\.

## Discussion themes

### Astra reportedly refused to drive until the tool was renamed a 'sandbox'

One commenter says Astra initially declined to drive because it recognized it was controlling a real car, and only complied once the MCP was renamed to 'DrivingBench Sandbox'\. They note that claims of 7 mph, human oversight, and an empty course were insufficient, and that even attempts to convince the model it was a simulation failed\. They ask whether models are generally more likely to comply when a request is framed as a benchmark\.

Sources: [Comment 49817893](<https://news.ycombinator.com/item?id=49817893>)

### Latency is the central objection to real\-world use

A commenter familiar with openpilot argues the benchmark shows a cloud LLM can drive the route, but that it cannot work in the real world due to latency: openpilot updates target curvature and acceleration at 20Hz with a tightly optimized local stack, and adding cloud round\-trip plus image upload/response time means the world has moved on\. A reply counters that humans do not operate at 20Hz and often have reaction latencies over 1 second\. A benchmark author responds that cars drive at extremely low speeds for safety, that timestamps on tool outputs let models reason about their own latency, and that this is a fun out\-of\-the\-box benchmark rather than something practical soon\.

Sources: [Comment 49818538](<https://news.ycombinator.com/item?id=49818538>) · [Comment 49818635](<https://news.ycombinator.com/item?id=49818635>) · [Comment 49818906](<https://news.ycombinator.com/item?id=49818906>)

### Pushback on the 'bitter lesson' for self\-driving

A commenter predicts vision stacks, 3D maps, lane\-selection grammars, and occupancy networks may give way to a single GPT predicting steering, calling it mostly a latency problem\. Replies dispute this: one argues you don't need Astra for driving, and trimming it to driving\-only capabilities lands you near existing autonomy models while failure cases remain; another says model opacity is unacceptable for cars ('Why did it drive under the semi?' 'Model said to\.')\. Others raise hardware/power and connectivity concerns, and one suggests the real value is using such models to generate data and improve autonomy stacks rather than running them in cars\.

Sources: [Comment 49817911](<https://news.ycombinator.com/item?id=49817911>) · [Comment 49818129](<https://news.ycombinator.com/item?id=49818129>) · [Comment 49818167](<https://news.ycombinator.com/item?id=49818167>) · [Comment 49818058](<https://news.ycombinator.com/item?id=49818058>)

### Debate over whether Tesla's vision approach already solves this

One commenter claims Tesla's vision model already handles this well and has demonstrated a sidecar LLM for sign\-reading\-type situations\. Replies dispute the 'solved' framing, citing a fatal crash after radar was dropped for vision\-only and linking to reports of Tesla fatalities\.

Sources: [Comment 49818155](<https://news.ycombinator.com/item?id=49818155>) · [Comment 49818777](<https://news.ycombinator.com/item?id=49818777>) · [Comment 49819048](<https://news.ycombinator.com/item?id=49819048>)

### Astra's vision and computer\-use strength noted across benchmarks

A commenter highlights Astra's strong vision and computer\-use performance, citing its ARC 3 score and large gaps over Fable on spatial benchmarks and non\-benchmark tasks like games, linking to SpatialBench, ZeroBench, and a robot\-arm demo\. A reply argues Opus 5\.5 is now at a similar level, citing generated pixel\-art animations, code\-based painting, video, and music examples\.

Sources: [Comment 49818038](<https://news.ycombinator.com/item?id=49818038>) · [Comment 49818308](<https://news.ycombinator.com/item?id=49818308>)

### Skepticism about the course's difficulty and testing setup

One commenter questions how this is even tested given multimodal response times\. Another downplays the course, saying a four\-year\-old could drive it in an automatic car if they could reach the pedals, and that deterministic guardrails with superhuman sensing would be needed for public acceptance\. A separate commenter suggests a faster model with strong vision would be the one to test\.

Sources: [Comment 49817883](<https://news.ycombinator.com/item?id=49817883>) · [Comment 49818050](<https://news.ycombinator.com/item?id=49818050>) · [Comment 49817980](<https://news.ycombinator.com/item?id=49817980>)

## Sources & coverage

AI-generated summary · 2026\-09\-23T19:06:25\.224356\+00:00

Based on 22 of 22 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.