"Lucasfilm and Disney have restored the original Star Wars (1977) and will release the film in multiple formats, including Imax 70mm [starting February 19], as part of the franchise's 50th anniversary celebrations," reports Deadline. The original film will also be displayed in its original 16:9 format on the center of Cosm's 12K+ LED Dome, in a special presentation developed with Disney, Lucasfilm, ILM, and MakeMake. They're calling it a "Shared Reality production" where "the surrounding visual environments dynamically evolve to immerse the audience in iconic Star Wars locations and moments." Deadline reports more 50th anniversary celebrations In addition to the Imax 70mm screenings, the film will also be presented in Digital Imax, Digital Cinema, 70mm, 35mm, and 4DX. On May 4th, 2027, Lucasfilm will release a new behind-the-scenes documentary, The Hero Awakens, from filmmaker R.J. Cutler on Disney+. Star Wars events already on the calendar for 2027 include Shawn Levy's [movie] Star Wars: Starfighter, which will be released May 28. There's also a new poster paying homage to the original 1977 film, reports the Disney news site Mickey Visit, plus a newly orchestrated score by John Williams that they assume "will be used on all the projects that debut this year." On April 2-3, [Disney] will debut the new "Star Wars in Concert: The Skywalker Saga" at the Hollywood Bowl. The show will be hosted by Anthony Daniels, known for his role as C-3PO. Special performances of John Williams' iconic film scores will feature scenes and music from all nine Skywalker Saga films. The concert is a sweeping celebration of the stories, themes, and music that have inspired audiences around the world for five decades... They also teased that new and legacy licensees — including Hasbro, the LEGO Group, adidas, Mars, Aston Martin, Steinway, Hajime Sorayama presented by Complex, and more — will introduce 50th-anniversary collections... In the opening of the press release, Disney mentioned that as part of the announcements there will be "reimagined Disney Parks and Cruise Line offerings." The article adds that the April's Star Wars Celebration at the Los Angeles Convention Center has already sold out. There's more Star Wars news, writes long-time Slashdot reader schwit1 A light saber used by Darth Vader in the "Star Wars" movies The Empire Strikes Back and Return of the Jedi sold Thursday for nearly $5 million in an auction. The light saber, which was used in the climactic duel in Cloud City in which Vader tells Luke Skywalker "I am your father" sold for $4.875 million at the Ultimate Star Wars Auction held by Heritage Auctions. The sale marks a world auction record for any "Star Wars" collectible, according to the auction house. The previous record was Heritage's $3,750,000 sale of Luke Skywalker's lightsaber and severed hand effects rig on July 15, 2026. And they're still auctioning more Star Wars memorabilia...

Read more of this story at Slashdot.

This article provides a roadmap for choosing where to run a model: on your own machine, through a hosted API, on Cloud Run, or on an accelerator you rent. Every figure comes from a measured run in gemma4-dev, which serves Gemma 4 on more than a hundred hardware and engine combinations and now carries five Gemini API models as sibling rigs.

https://github.com/xbill9/gemma4-dev

The New Gemini Models in One Table

Gemini 4 Hydrogen, Carbon and Argon arrived on the Gemini API next to 3.5 Flash-Lite, 3.8 Flash and Gemma 4. On 2026-11-14 all seven ran through the same Gemini API endpoint, started together, with the same GSM8K (1,319 math word problems) and BFCL simple (400 function calls) questions and scoring, with thinking set as low as each model accepts:

Hydrogen, Carbon and Argon are ahead of Gemma 4 31B on GSM8K, with the 95% range of the difference above zero, and level with it on BFCL. Output prices include thinking tokens, and thinking decides the bill. On Hydrogen and 3.8 Flash it cannot be turned off: Hydrogen has the lowest Gemini 4 price, thought on every problem at its lowest setting, and cost the most. It is also the fastest Gemini 4 model, about 2.7 times quicker per problem than Argon. The times come from all seven sharing one connection at once. Carbon and Argon double in price on 2026-12-01. The method is in the companion article, Gemini 4 Hydrogen, Carbon and Argon Against Gemma 4.

The Options

Four ways to run a model, each measured in this repo:

• Local: a GPU or CPU you own. Measured on an RTX 3050 OEM (8 GB desktop card), a GTX 1650 Ti (4 GB laptop card) and an Intel i7-10750H laptop CPU.
• Hosted API: someone else's server, reached with a key. Measured on the Gemini API (the five Gemini models) and AMD Token Lab (Gemma 4 31B, free with the AMD AI Developer Program).
• Cloud Run: Google's serverless containers with one NVIDIA L4, serving Gemma 4 E2B on vLLM.
• Your own accelerator: a rented machine you start and stop. Measured on a TPU v5e (one chip and four), an EC2 G6 with one L4, and an AMD MI300X on DigitalOcean.

Option 1 — Local

The RTX 3050 serves Gemma 4 E2B on six engines. One request at a time without a drafter they run from 61.9 tok/s (LiteRT-LM) to 121.1 tok/s (vLLM); with Google's speculative-decoding drafter vLLM reaches 230.5 tok/s. Under load, with 64 requests at once of 128 tokens in and 512 out, SGLang delivers 1,711.3 tok/s in total and vLLM 1,651.2.

The same card holds the larger models one request at a time: 12B at 56.4 tok/s with its drafter, 26B A4B at 40.6 tok/s on code with its experts in system RAM, and 31B at 4.0 tok/s with two thirds of it on the CPU. Quality rises with size: GSM8K 0.9014 (E2B), 0.9629 (12B), 0.9735 (31B).

The 1650 Ti laptop card decodes E2B at 73.75 tok/s for one request and peaks at 164 tok/s through its server at 16 requests. The laptop CPU decodes 16.71 to 18.86 tok/s.

Setup: llama.cpp is ready 4 s after start, vLLM 48 s. vLLM's first start compiles for about 3.5 minutes. The hardware is already paid for, and nothing is billed per hour.

Option 2 — Hosted API

Gemini API. A key and an HTTP call. The Gemini models and Gemma 4 26B and 31B answer the OpenAI chat format at /v1beta/openai/chat/completions. Estimated from the token counts returned and list prices, a full 1,319-problem GSM8K run cost $0.87 on Flash-Lite, $1.66 on 3.8 Flash, $3.15 on Argon, $3.70 on Carbon and $4.51 on Hydrogen. Gemma 4 is on the free tier only, with paid use not available, and its free-tier rate limit answered three requests in the GSM8K and BFCL runs with HTTP 429 after every retry.

AMD Token Lab. Gemma 4 31B on an SGLang router on AMD GPUs, free with the AMD AI Developer Program: 10 points for 30 days, with Gemma 4 31B at 0.035 points per million input tokens and 0.07 per million output. One request at a time decoded at 2.0 to 7.6 tok/s on the shared service, with up to 5 s before the first token. The limit is 128 requests at once across all users; over that, the reply is HTTP 429.

Option 3 — Cloud Run

One L4 GPU in us-east4 with 8 vCPU and 32 GiB, Gemma 4 E2B on vLLM. Output rose with concurrency to 188.46 tok/s at 4 requests and stayed there at 8 (189.53), because the service was deployed with --concurrency=4; more users means more instances. The same L4 on EC2 reached 360.17 tok/s at 8.

At list price the instance costs $1.4209 an hour, which works out to $7.95 per million output tokens at one request and $2.08 at eight. Cloud Run scales to zero by default, so an idle service costs nothing and the next request waits for a cold GPU start; the startup check waits 180 s before its first look. The weights load from Cloud Storage.

Option 4 — Your Own Accelerator

TPU v5e, one chip. Gemma 4 E2B on vLLM: 143.9 tok/s for one request, 1,689.1 tok/s at 64 requests (128 in, 512 out), GSM8K 0.9007 and BFCL 0.920. Flex-start costs $0.60 per chip-hour, about $0.089 per million output tokens at full batching and $1.31 at one request. Capacity came within two to five minutes in us-west4-a, and the engine took 12.6 to 14.1 minutes to start on most rigs, most of it compiling (27.2 on one).

TPU v5e, four chips. E2B as four one-chip engines reaches 7,788 tok/s at 64 requests and 14,189 at 256. The 12B fits with 122,624 tokens of context cache, against 9,728 on one chip.

EC2 G6, one L4. vLLM at 132.8 to 140.4 tok/s for one request and 3,715.7 tok/s at 64; with the drafter, 4,516.3. A spot g6.2xlarge measured on an older vLLM cost $0.9412 an hour, $0.73 per million output tokens at eight requests.

AMD MI300X. 192 GB on one card: E2B at 340.4 tok/s for one request and 10,293.6 tok/s at 64 (128 tokens of context). $1.99 an hour, billed while powered off too: $0.054 per million output tokens at 64 clients and $1.624 for a single stream.

🔎 Tip: Price Per Hour Becomes Price Per Token Only Under Load

Every rented option above has two prices per million tokens, one at a single request and one at full batching: Cloud Run $7.95 and $2.08, TPU v5e $1.31 and $0.089, MI300X $1.624 and $0.054. A machine that serves one user at a time costs about what a hosted API charges, or more. A per-token API charges the same at any load.

Compare and Contrast

The Roadmap

Pick by the axis that matters most, then check the others.

By number of users.

By cost. Free first: your own card, Token Lab, and the Gemini free tier, which carries Gemma 4 26B and 31B, Flash-Lite, 3.8 Flash and Hydrogen. For paid use with light or uneven traffic, pay per token: both tests cost $0.92 on Flash-Lite and $3.46 on Argon. Hydrogen thought on every request and 3.8 Flash on most; Google's docs say reasoning cannot be turned off on Gemini 3 or Gemini 4 models, and the API runs none as low. Compare models by cost per answer. From 2026-12-01, when Carbon and Argon go to $4 / $20, this run costs $6.93 on Argon and $5.02 on Hydrogen, by arithmetic from list prices. A rented machine wins only when it is kept busy, which brings TPU v5e to $0.089 and MI300X to $0.054 per million output tokens.

By speed. On a hosted API, Flash-Lite answered a GSM8K problem in 1.10 s and Hydrogen in 2.84 s; Gemma 4 on the Gemini API took 8 to 9 s. For one user, a local card with a drafter (230.5 tok/s on the 3050) or an MI300X (340.4). For many users, the MI300X or four TPU v5e chips. Cloud Run as deployed here stops at 189.53 tok/s per instance.

By accuracy. Gemini 4 Argon leads GSM8K at 0.9841, and Argon and Gemma 4 31B lead BFCL at 0.9675, on short problems with thinking at its lowest setting. For hard, multi-step work Google calls Carbon "our strongest frontier model"; that case was not measured here. Where a model runs did not change its score: Gemma 4 31B answered 1,284 of 1,319 GSM8K problems through the Gemini API and 1,284 on an RTX 3050 with llama.cpp, differing on 8 problems. Model size matters more than where it runs: E2B scored 0.8954 to 0.9030 on GSM8K across six engines on the RTX 3050, the L4 and TPU v5e, and the 31B 0.9735.

By ease of use. A hosted API needs a key. A local card needs a driver and an engine, and llama.cpp is ready in seconds. Cloud Run needs a container and a deploy and then scales on its own. A rented accelerator needs capacity, a 12.6 to 14.1 minute engine start on TPU, and someone to turn it off.

So, Which One?

Start hosted with Gemma 4 31B on the Gemini free tier: it tied the best BFCL score and costs nothing. If the task needs the best math scores, move to Gemini 4 Argon. If the task needs Gemma 4 for its open weights, privacy or offline use, run it on your own card, 12B or 31B on 8 GB. Move to Cloud Run when a few users need an always-reachable endpoint without managing a machine. Move to a TPU v5e or an MI300X when traffic is steady enough to keep it busy, because that is where the per-token cost falls below every API.

Summary

The goal of this article was to map where each way of running a model fits. The key to the solution was a single repo that measures local cards, hosted APIs, Cloud Run and rented accelerators with the same engines, scoring script and questions. The results were:

• 🟢 Gemini 4 Argon has the best GSM8K score measured, 0.9841; Argon and Gemma 4 31B lead BFCL at 0.9675.
• 🟢 Gemma 4 31B scored the same 1,284 of 1,319 on GSM8K through the Gemini API and on an 8 GB RTX 3050.
• 🟢 Gemma 4 31B is free on the Gemini API and tied Argon on BFCL; Argon cost $3.46 for both tests.
• 🟢 A busy MI300X or TPU v5e serves Gemma 4 E2B at $0.054 and $0.089 per million output tokens.
• 🟢 AMD Token Lab and the Gemini free tier cost nothing to start.
• ⚠️ A rented machine at one request costs $1.31 (TPU v5e) to $7.95 (Cloud Run) per million output tokens.
• ⚠️ Cloud Run as deployed caps one instance at 4 requests, 189.53 tok/s.
• ❌ Token Lab's shared endpoint decoded one request at 2.0 to 7.6 tok/s.

Scope: figures come from separate runs between 2026-09-03 and 2026-11-14 in gemma4-dev, on different engine versions and days; each is quoted from its run's report, and ratios between runs on different hardware are indicative. Prices are list prices on the date each run recorded them. Accuracy was measured on GSM8K and BFCL simple only, and not on Cloud Run, Token Lab or the MI300X; the throughput figures are Gemma 4 E2B unless a size is named. Gemini API throughput under load was not measured.

The strategy for choosing where to run Gemma 4 and Gemini was validated with an incremental step by step approach.

References

• gemma4-dev on GitHub
• Evidence for this article
• Gemini API pricing
• AMD Token Lab
• Cloud Run GPUs
• Cloud TPU
• vLLM
• llama.cpp
• Gemma 4 on Hugging Face

Hope you enjoyed it!

Anthropic is positioning its new model, Claude Fable 5.2, for demanding coding and technical work. Anthropic launched Fable 5.2 on Nov. 9 and reported higher scores than OpenAI’s GPT-6 Astra on two coding benchmarks. Fable 5.2 also has lower standard API rates.

Anthropic’s launch benchmarks put Fable 5.2 ahead of Astra, 49.1% to 42.8% on FrontierCode 2.0 and 69.9% to 65.5% on FrontierSWE v2. Fable’s 49.1% came at X high effort. At High, its default on the Claude API, it scored 45.3%. Benchmark leads don’t always carry over to real work.

The cost side is less simple than the price list. Tokens are the chunks of text a model reads and writes, and AI companies bill by them. At standard API rates, Fable 5.2 costs $8 per million uncached input tokens and $40 per million output tokens. OpenAI launched GPT-6 Astra in September at standard API rates of $10 per million uncached input tokens and $50 per million output tokens. “The price per task is what matters,” OpenAI President Greg Brockman said at Astra’s launch. Both APIs charge for reasoning tokens as output and have separate rates for cached input.

In my experience, Anthropic’s models have used more tokens than OpenAI’s. I put this positioning to the test and ran Fable 5.2 against Astra on tasks that simulate real developer work.

The tests

I ran both models on the same three tasks, once each, through each vendor’s API at default settings. Each test is worth 50 points, and I used a script to grade all the answers. Here are the tests:

• Bug fix: I planted 10 bugs in a 2,600-line billing codebase and wrote a support ticket describing 10 customer complaints. The ticket’s guesses about the causes are wrong, two of the codebase’s own tests are wrong, and one bug only shows up after another is fixed.
• Code review: I planted 45 defects in its VAT ID checks for 27 countries, its IBAN rules, its European Central Bank payment calendar, and its date formats for 50 locales in an EU invoicing package. The models had to prove each defect with an input the code gets wrong and identify the correct answer. Flagging false alarms cost the models points.
• Code writing: The model wrote the same helpers from scratch, including VAT ID validation for 27 countries, IBAN checks for 34 countries, the payment calendar, and long dates.

Bug fix

Fable 5.2 and Astra fixed all 10 bugs and passed all 50 hidden checks, both in 12 turns (the matching number of turns surprised me).

The difference is in how much each model spent to get there. Fable 5.2 used more than 17,000 reasoning tokens, and Astra used 819 (wow). Astra passed the checks in about two minutes at an estimated cost of 98 cents. Fable 5.2 took more than five minutes and cost an estimated $2.51.

The numbers:

• Fable 5.2: 12 turns, 696,863 in, 36,135 out (17,387 reasoning), $2.51, 326 s.
• Astra: 12 turns, 231,854 in, 6,823 out (819 reasoning), $0.98, 122 s.

Code review

This is the only test where the scores differ. Fable 5.2 scored 49, and Astra scored 45. Fable 5.2 proved 44 of the 45 defects and raised no false alarms across 71 findings. Astra proved 41 and incorrectly classified one Spanish ID as valid. Astra missed defects in check-digit rules for Spain, Ireland, and Luxembourg. Both models missed the same Spanish defect: two swapped letters in a 23-letter lookup table.

Fable 5.2’s thoroughness came at a price. It cost more than six times what Astra did and took more than four times as long. If your code handles invoices (or simple approvals), one validation rule the review misses can get real invoices rejected, so Fable 5.2’s four extra points may be worth the higher price. For a routine review, Astra found 91% of the defects for 16% of the cost. Each model ran once, so a four-point gap could shift on another run.

The numbers:

• Fable 5.2: 26 turns, 2,207,564 in, 103,240 out (77,641 reasoning), $6.05, 993 s.
• Astra: 12 turns, 192,018 in, 8,664 out (4,307 reasoning), $0.95, 227 s.

Code writing

Both models scored 50 of 50. Each one wrote working checks for 27 countries’ VAT IDs and 34 countries’ IBANs with no libraries to lean on. Neither model had trouble here. Both passed all the checks in one run.

Astra took 23 turns, almost as many as Fable 5.2’s 24. Fable 5.2 still cost 2.3 times as much and took 1.7 times as long for the same score. In this run, the extra tokens didn’t raise the score.

The numbers:

• Fable 5.2: 24 turns, 1,495,712 in, 87,542 out (47,458 reasoning), $4.83, 838 s.
• Astra: 23 turns, 460,271 in, 24,007 out (9,328 reasoning), $2.06, 483 s.

Results

Test	Claude Fable 5.2	GPT-6 Astra
Bug fix (of 50)	50	50
Code review (of 50)	49	45
Code writing (out of 50)	50	50
Total tokens in / out	4,400,139 / 226,917	884,143 / 39,494
Estimated cost	$13.39	$3.99
Total time	2,157 s	832 s

Fable 5.2 led on code review and tied Astra on the other two tasks. Across the two tied tasks, its estimated API cost was 2.4 times Astra’s ($7.34 vs. $3.04), and it took 1.9 times as long (1,164 vs. 605 seconds).

What do I think?

For tasks like these, I’d start with Astra. It matched Fable 5.2’s scores on bug fixing and code writing at about 41% of the estimated cost, and it finished all three tasks faster. I’d consider Fable 5.2 for a careful code review, where it found three more defects in this test. Both models missed defects.

Fable 5.2’s standard token rates are lower, but it used about five times as many tokens as Astra across these three runs. Brockman’s point about price per task held up here.

The post Claude Fable 5.2 vs. GPT-6 Astra: A lower list price didn’t mean a lower bill appeared first on The New Stack.

Ben Nadel looks at enabling TLS 1.2 handshakes for the CFHttp tag running on CommandBox and Adobe ColdFusion....

You likely chose most of the changes in your career yourself, and many engineers now feel AI-driven change is being forced on them. In this episode, I talk about why that shift feels so personal, and how three ideas can help you face it: the Stoic principle that "the obstacle is the way," Nassim Taleb's antifragility, and the biological cycle of stress and adaptation.

For most of my career, the changes I went through happened at my own pace. I picked what to learn, whether that was jQuery animations or the next framework, and each new skill felt like a superpower. Many engineers today will recognize that progression, and it's a big part of why we love this work. Now the AI-driven shift in our industry is changing that for a lot of people. The change isn't chosen anymore, and for many engineers it's scary. In today's episode, I talk about why this shift feels so personal, and I share a framework you can use to grow through it instead of avoiding it.

• Change You Chose vs. Change Chosen for You: A lot of what made software engineering rewarding was agency. We picked our tools, our learning paths, and often our jobs. Agentic engineering is taking away some of that choice, and it's worth naming that loss honestly.
• A Note for Engineering Managers: If you never went through this kind of progression as an engineer, this part is for you. Code is more than a convenient tool for your engineers. It's something many of them love. Like a blacksmith with a hammer they've used for 20 years, they may not want the "better" tool, and a purely rational argument about better tooling misses the human side of the conversation.
• The Obstacle Is the Way: Drawing from Ryan Holiday's book and the Stoic philosophy of Marcus Aurelius it's based on, consider treating the obstacle in front of you as the path forward instead of something standing in your way.
• Antifragility and Stress Adaptation: Nassim Taleb's Antifragile describes systems that are actually improved by stressors. The same pattern shows up in biology, as in strength training: stress, adaptation, recovery, repeat. With proper recovery, the stressor leaves you more resilient to it.
• Why Avoidance Makes It Worse: Cognitive behavioral therapy shows that avoiding what we fear tends to feed rumination. Safely and rationally confronting that fear, and learning that you survived it, is what drives the adaptation.
• An Integrated Path Forward: If you want to stay in this industry and keep building software, you don't have to give up what you loved about the work. You can face the thing you're afraid of, give yourself time to recover, and then go again.
• Episode Homework: Find a quiet space and reflect honestly. Ask yourself, "What about this industry shift am I avoiding?" There's no right answer, but what you're avoiding may be the most useful thing for you to confront.

📮 Ask a Question

If you enjoyed this episode and would like me to discuss a question that you have on the show, drop it over at: developertea.com.

📮 Join the Discord

If you want to be a part of a supportive community of engineers (non-engineers welcome!) working to improve their lives and careers, join us on the Developer Tea Discord community today!

🗞️ Subscribe to The Tea Break

We are developing a brand new newsletter called The Tea Break! You can be the first in line to receive it by entering your email directly over at developertea.com.

🧡 Leave a Review

If you're enjoying the show and want to support the content head over to iTunes and leave a review!

Download audio: https://dts.podtrac.com/redirect.mp3/cdn.simplecast.com/audio/c44db111-b60d-436e-ab63-38c7c3402406/episodes/dd39199f-ea5a-448d-bdd0-85177b2b2fb7/audio/d128ca50-f368-47f7-a89f-df02416d7276/default_tc.mp3?aid=rss_feed&feed=dLRotFGk

---