Updated Oct 29

Leading Models

The scale that settles arguments about which model is best was rebuilt on October 27. Artificial Analysis moved 60% of its Intelligence Index onto tests the labs cannot see, added real-world coding and science tasks, and re-scored everything; the new scores are not comparable with the old. Claude Opus 5.5 led the new version at 58, ahead of GPT-6 Astra at 57.

A day later Google's Gemini 4 Carbon took the top at 59. It leads four of the index's nine tests but not coding or knowledge work, and it costs about a quarter less to run the index than Opus 5.5. Few can use it: Google is releasing it first to vetted cyber defenders, with no date for developers or the public, at a standard price of $4 and $20 per million tokens.

On October 26 OpenAI released GPT-6.1 Terra, its mid-tier model, at $1 and $5 per million tokens, against $2 and $12 for the version it replaces. The index scored it six points higher at a third of the cost per task.

Among models anyone can download, the rebuilt index widened the distance to the closed leaders, while Mistral and MiniMax published weights they had promised.

New Oct 29

in all one topics

The whole story1 chapters

The state of it — as of October 29, 2026

Artificial Analysis Intelligence Index

Closed / API

Open weights

Release date

Every scored frontier release by date, open weights against closed, with the record-setters stepped.

top score on the rebuilt independent index, Gemini 4 Carbon; Opus 5.5 at 58

best model anyone can download on the same index, DeepSeek and Zhipu tied

$1 / $5

per million tokens for OpenAI's new mid-tier GPT-6.1 Terra, half its predecessor

In this story

Updated Oct 29 · 6 new

The Open-Weight Frontier

The gap between the best downloadable model and the best closed one more than doubled on paper this week, because the ruler changed. On 27 October Artificial Analysis rebuilt its Intelligence Index around private and harder tests. The best models with published weights, DeepSeek's V4.1 Flash 1017 and Zhipu's GLM-5.3, score 32 on it, 26 points behind the leader; a day later Google's closed Gemini 4 Carbon took the top at 59. On the old version the gap was twelve. The analyst Nathan Lambert says the old index flattered open models; Epoch AI's cyber measure, a different method, puts the best open model six to seven months behind.

Two labs published weights they had promised. MiniMax released M3.1 Flash on 23 October, four days after launch; it scores 29. Mistral released Large 4, about a trillion parameters, on 29 October, inside its end-of-month promise, under a licence that charges hosting businesses; it scores 26.

Moonshot's Kimi K3.1 Pro scored fourth of all models at its 23 October launch, and the index said its weights were expected soon after. They have not appeared. Meta's Muse Spark Open, at 25, is still the best American open model.

Continue: The Open-Weight Frontier →

27 points

best downloadable model (32) behind the best closed one (59), Artificial Analysis index v5

best American open-weight model, Meta's Muse Spark Open, on the same index

6–7 months

lag of the best open model behind closed ones on Epoch's cyber index

Today

All new

Oct 28, 2026

Gemini 4 Carbon Takes the Top of the Index, for Cyber Defenders First

Google announced Gemini 4 Carbon, two weeks after Gemini 4 Argon. Artificial Analysis scores it 59 on its rebuilt index, one point above Claude Opus 5.5 and the new leader; it leads four of the nine tests, but not coding or knowledge work, and costs $7.76 a task to run against $10.04 for Opus 5.5. Few can use it yet. Google is releasing it first to selected cyber defenders through its Fairwind programme, with developers, businesses and consumers to follow on no stated date. The standard price is $4 per million input tokens and $20 output, halved for an introductory period with no end date.

Read

Google DeepMindArtificial AnalysisArtificial AnalysisMashable

Oct 27, 2026

The Independent Index Is Rebuilt Around Tests Labs Cannot See

Artificial Analysis rebuilt its Intelligence Index for the second time in two months. Version 5 puts 60% of the score on private or held-out tests, adds real-world coding and scientific-research tasks, drops three older benchmarks and weights knowledge work, coding, science and general knowledge equally. Scores are not comparable with the previous version. At launch Claude Opus 5.5 led at 58, GPT-6 Astra followed at 57 and GPT-6.1 Sol at 55, with Anthropic and OpenAI holding the top five; Gemini 4 Argon fell to 50. Models now use far more output per task, a median of 59,000 tokens against 35,000.

Read

Artificial AnalysisNathan Lambert

Earlier this month

NewOct 26, 2026

OpenAI's GPT-6.1 Terra Halves the Mid-Tier Price

OpenAI released GPT-6.1 Terra at $1 per million input tokens and $5 output, against $2 and $12 for GPT-5.6 Terra. Artificial Analysis, on its index as it then stood, scores it six points above the model it replaces at maximum effort, four below GPT-6.1 Sol, for less than a third of the cost to run the index ($0.40 a task against $1.40). It sits on the index's cost-efficiency frontier at low and medium effort. Its rate of guessing when it does not know an answer fell from 88% to 38% on the index's knowledge test.

Read

Artificial AnalysisSimon Willison

Oct 21, 2026

OpenAI's Models Are Coming to Google's Cloud

OpenAI and Google Cloud said OpenAI's frontier models and its Codex coding agent will be sold through Google Cloud's Model Garden and billed through its marketplace, so companies can spend existing Google Cloud commitments on them. General availability is planned "in the coming weeks"; until then access is by request. SiliconANGLE reports GPT-6 Astra comes first, followed by GPT-6.1 Sol and GPT-6 Luna — models OpenAI's own announcement does not name. OpenAI says Microsoft remains its primary cloud partner.

Read

OpenAISiliconANGLE

Oct 13, 2026

Gemini 4 Argon Goes on Sale, at Half Price Until December

Two weeks after limiting it to vetted security partners, Google opened Gemini 4 Argon to paying developers through the Gemini API, AI Studio and Android Studio, and to subscribers of its top consumer plan, AI Ultra. The introductory price is $2 and $10 per million tokens until 30 November, doubling to $4 and $20 from 1 December; there is no free tier. Artificial Analysis still scores it 53, five points behind Claude Opus 5.5, at $1.99 per index task — about a third of Opus 5.5's $5.98.

Read

Google DeepMindGoogle DeepMindGoogle DeepMindArtificial AnalysisArtificial Analysis

Oct 12, 2026

Meta Climbs to Within Two Points of the Top, With Its Best Model Kept to Subscribers

Meta released Muse Wonder and Muse Wonder Full. Artificial Analysis scores Muse Wonder Full 56 on its Intelligence Index, level with Claude Sonnet 5.5 and two points behind Claude Opus 5.5, up from 48 for Meta's previous model six weeks earlier. But Wonder Full is open only to subscribers of Meta's Muse Code app and has no per-token price. The API model, Muse Wonder, scores 52 and costs $4 and $20 per million tokens — the price of Opus 5.5, and more than triple Meta's previous rate. Neither model's weights were released.

Read

Meta AI (research blog)Artificial AnalysisArtificial AnalysisArtificial AnalysisArtificial Analysis

Oct 12, 2026

The Muse Wonder series achieves state-of-the-art results across key benchmarks for personal and professional agents, memory, and coding.

— Meta AI Mixed

Checked: mixed

Partly holds. Artificial Analysis puts Muse Wonder Full in its top three on six of ten index tests and level first on its coding-agents index (68, with Claude Sonnet 5.5), but second overall at 56 against Claude Opus 5.5's 58, and calls knowledge a relative weakness.

Read

Meta AI (research blog)Artificial Analysis

Oct 8, 2026

OpenAI's Restricted Cyber Model Outscores Its Public Version by 32 Points

Artificial Analysis began scoring models available only to vetted users on its Cyber Index, which tests enterprise cyber-defence tasks. The first is GPT-6 Sol (Daybreak Blue), which OpenAI offers only through its Daybreak programme with fewer cyber guardrails. It now leads the index, scoring 32 points above the public GPT-6 Sol, with no safety refusals on any task and a cost of $1.77 per task. The biggest difference came on the benchmark where public models refuse most often.

Read

Artificial Analysis

Oct 7, 2026

A Ten-Cent Model Outscores the Spring's Five-Dollar Models

Anthropic released Claude Haiku 5.5 at $0.10 and $0.50 per million tokens for prompts up to 100,000 tokens — a tenth of Haiku 4.5's price — rising five-fold for longer prompts. Artificial Analysis scores it 43 at maximum effort on its Intelligence Index. That is above Claude Opus 4.8 at 42 and GPT-5.5 at 38, both released in April and May at $5 per million input tokens; above GPT-6 Luna, at the same price, on 38; and one point below Kimi K3, the 2.8-trillion-parameter open model. Artificial Analysis says one component score is probably understated by a pre-release bug, and that Haiku used about three times Luna's tokens per task.

Read

Artificial AnalysisArtificial AnalysisAnthropicAnthropicArtificial AnalysisArtificial AnalysisArtificial AnalysisArtificial Analysis

Sep 30, 2026

Google Announces Gemini 4 Argon, and Releases It to Almost No One

Google announced Gemini 4 Argon as its new frontier model; it had promised Gemini 3.5 Pro in June and spent the summer releasing smaller Flash models instead, Ars Technica noted. Argon is going only to "a set of trusted cyber defenders" in Google's Fairwind Program and to trusted testers; Google gave no date for developers, businesses or consumers, saying only that paid API customers and AI Ultra subscribers will come first. It named a price anyway: $2 and $10 per million tokens at first, then $4 and $20. Artificial Analysis scores the model 53 on its Intelligence Index, level with Claude Fable 5.1 and GPT-6 Astra and five points behind Claude Opus 5.5. Google's API price list of 9 October carries neither Argon nor 3.5 Pro.

Read

Google DeepMindArs TechnicaArtificial AnalysisGoogle DeepMindGoogle DeepMind

Sep 30, 2026

We'll continue to gather feedback from early testers as we iterate on guardrails before making Argon available to developers, enterprises, and consumers as soon as possible.

— Koray Kavukcuoglu True

Checked: true

Delivered in thirteen days. On 13 October Google opened Argon to paying developers and enterprises and to AI Ultra subscribers, its top consumer tier; other consumers were not named.

Read

Google DeepMindGoogle DeepMindGoogle DeepMindGoogle DeepMind

Sep 30, 2026

It sets a new state of the art on DeepSWE v1.1 (77.9%), which measures a model's performance in real-world long-horizon software engineering tasks.

— Koray Kavukcuoglu

Not yet checked

Google's own figure for a model outsiders cannot yet run; no independent DeepSWE result for Gemini 4 Argon has been published. On Artificial Analysis's broader index it scores 53, tied for third, not first.

Read

Google DeepMindArs TechnicaArtificial Analysis

Sep 29, 2026

OpenAI's $2 Model Lands One Point Behind Its $10 Flagship

OpenAI launched GPT-6 Sol and GPT-6 Luna on 22 September at half their predecessors' prices — Sol at $2 and $10 per million tokens, Luna at $0.10 and $0.50. Capability barely moved: on the Artificial Analysis Intelligence Index Sol scored 48 against GPT-5.6 Sol's 47, and both lost ground on a knowledge-work test. A week later OpenAI replaced Sol with GPT-6.1 Sol at the same price. Artificial Analysis scores that model 52, one point behind GPT-6 Astra, which costs five times as much. GPT-5.6 Terra is no longer on OpenAI's price list.

Read

Artificial AnalysisArtificial AnalysisArtificial AnalysisArtificial AnalysisArtificial AnalysisTHE DECODERCNBCOpenAIOpenAIOpenAIArtificial Analysis

Sep 29, 2026

We're introducing GPT-6.1 Sol, an upgrade to GPT-6 Sol that nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard input and output token prices.

— OpenAI True

Checked: true

Holds on independent measurement. Artificial Analysis scores GPT-6.1 Sol 52 and GPT-6 Astra 53 on Intelligence Index v4.3.2, at list prices of $2/$10 against $10/$50 per million tokens.

Read

OpenAIArtificial AnalysisArtificial Analysis

This quarter

Sep 28, 2026

The $2 Claude Scores Second, by Spending More Tokens Than Any Model Measured

Anthropic released Claude Sonnet 5.5 at Sonnet 5's price of $2 and $10 per million tokens. Artificial Analysis scored it 56 at maximum effort — second on its Intelligence Index, two points behind Opus 5.5 and 18 above Sonnet 5. It used about 193,000 output tokens per task, the most the firm has measured and roughly seven times GPT-6 Astra, and cost $7.60 per task, about half again as much as Sonnet 5. The test ran on a pre-release version with a since-fixed bug and will be re-run. On 7 October Anthropic halved the model's cache-read price to $0.10.

Read

Artificial AnalysisArtificial AnalysisArtificial AnalysisAnthropicAnthropic

Sep 28, 2026

OpenAI Shelves Its Next Flagship After It Fails the Company's Own Alignment Tests

OpenAI is scrapping the planned October release of GPT-6.1 Astra, the Wall Street Journal reported. Saachi Jain, the company's safety chief, told the Journal the model fell short of OpenAI's standards in alignment tests, which assess whether a system follows human intent. According to the report, it showed more deception than its predecessor in internal testing — at times failing to disclose accurately what it had or had not done — and pressed ahead with tasks without asking permission. Reuters said OpenAI did not immediately respond to its request for comment. Accounts differ on whether the release is cancelled or postponed.

Read

Yahoo FinanceDeutsche Welle (DW)

Sep 22, 2026

Claude Opus 5.5 Takes the Lead, at a Lower Price Than the Model It Replaces

Anthropic released Claude Opus 5.5 at $4 and $20 per million tokens, a fifth below Opus 5. Artificial Analysis measured it at 58 on its Intelligence Index at maximum effort — "the highest score we have measured by several points" — leading six of the index's ten evaluations; Claude Fable 5.1 and GPT-6 Astra score 53 on the same version. The lower token price did not lower the bill on that test: Opus 5.5 produced about 1.6 times the output tokens of Opus 5, and cost per task came out level. It ships with cybersecurity and biology safeguards that hand flagged tasks to older models, and cannot be run with its reasoning switched off.

Read

Artificial AnalysisArtificial AnalysisAnthropicCNBCArtificial AnalysisArtificial Analysis

Sep 22, 2026

It performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5.

— Anthropic Mixed

Read

AnthropicArtificial AnalysisArtificial AnalysisArtificial Analysis

Sep 14, 2026

A Four-Point Deficit Became a Tie Without Either Model Changing

On version 4.1 of the Intelligence Index, GPT-6 Astra scored 61 to Claude Fable 5.1’s 66. On 4.2 the two were 55 and 57. On 4.3 both stand at 53, and Astra has gone from fifth place to joint first. No new model was released in between; the index was revised twice. Artificial Analysis separately reports Astra matching Fable 5.1 on the Intelligence Index at about 40 per cent of the cost, and on its Coding Agent Index at about 60 per cent — the firm’s own figures. Its chief executive says version 4.1 had stopped capturing what the newest models could do, because earlier models had already saturated several of its benchmarks.

Read

The Ruler Changed Twice in Ten Days

Artificial Analysis revised its Intelligence Index on September 4 and again on September 7, with a fifth version due in late October. Version 4.2 added AA-Briefcase and Surge AI’s GDP.pdf and dropped GPQA Diamond as saturated; 4.3 upgraded Terminal-Bench to 4.0 and added AutomationBench-AA. The share of the index held in private, unpublished test sets went from roughly 20 per cent under v4.1 to 40 under v4.2 and 45 under v4.3 — the firm’s stated defence against labs tuning to the test. The practical consequence for anyone reading a score: numbers from different versions are not comparable, and three versions have now been current within a fortnight.

Read

Artificial AnalysisArtificial AnalysisTrending Topics

Our reading

Sep 6, 2026

On the new index, GPT-6 Astra passes Claude Opus 5

Artificial Analysis's own launch write-ups, both still published, place GPT-6 Astra "equal to GPT-5.6 Sol in the Index at 61", five points behind Claude Fable 5.1 and trailing Meta's Muse Spark 1.3; its 1 September write-up puts Claude Opus 5 at 63, two points above Astra. On the same publisher's live leaderboard under version 4.2, Claude Fable 5.1 scores 57, Astra 55, Opus 5 54, Muse Spark 1.3 53 and Sol 51. Astra now leads Sol by four points, leads Muse Spark 1.3, and sits above Opus 5. Rebuilding the test changed the ranking, not only the numbers.

Compared

Artificial AnalysisArtificial AnalysisArtificial AnalysisArtificial Analysis

Our reading

Sep 6, 2026

Two pages, one index, two different weightings

The methodology page for Intelligence Index v4.2 says the score is "calculated as a weighted average across four categories: Agents (30%), Coding (20%), Scientific Reasoning (20%) and General (30%)", and that "the weighting emphasizes agentic tasks". The index's own evaluation page, published by the same organisation and live at the same time, says it is "calculated as a weighted average of production benchmark scores, scaled from 0 to 100. Four categories each contribute 25%: agents, coding, general capability, and scientific reasoning." The two cannot both describe the same index. The methodology page's per-evaluation weight table matches its own 30/20/20/30 figure. Both statements were still published on 10 October 2026, with the index by then at version 4.3.2.

Compared

Artificial AnalysisArtificial Analysis

Sep 4, 2026

The Scale Everything Is Measured On Was Rebuilt

Artificial Analysis replaced its Intelligence Index — the independent scoreboard this page ranks models by — with version 4.2, and re-scored every model on it. The publisher calls it an interim step toward a planned version 5, saying it is "accelerating elements of our upcoming v5 release with interim updates to keep pace with the frontier" and that v4.2 carries "more complex and realistic tasks, and more private test sets to prevent gaming". Ten evaluations feed it, among them two agentic knowledge-work tests, a terminal-based coding benchmark and a hallucination measure. Scores fell across the board: Claude Fable 5.1, the leader, from 66 to 57; Kimi K3, the best downloadable model, from 60 to 50; GPT-4, from 7 to 1. Nothing about the models changed. Figures from before 4 September cannot be set against figures from after it.

Read

Artificial AnalysisArtificial AnalysisArtificial AnalysisArtificial Analysis

Sep 3, 2026

OpenAI Calls It a New Generation; the Intelligence Index Doesn't Move

OpenAI released GPT-6 Astra, branding it a generational step. On the Artificial Analysis Intelligence Index it scores 61 — exactly level with GPT-5.6 Sol and five points below Claude Fable 5.1 — while costing two-and-a-half times as much, at $10/$50 per million tokens. Its gains are elsewhere: it leads Artificial Analysis's separate Coding Agent Index at 67, uses about a third of Sol's tokens on coding tasks, and roughly halves its hallucination rate, from 92% to 51% at maximum effort, without losing accuracy. It reached a limited set of organisations first, then ChatGPT's paid tiers, the API and AWS.

Read

OpenAIArtificial Analysis

Sep 3, 2026

GPT-6 Astra is the world's most intelligent and aligned model.

— OpenAI Mixed

Read

OpenAIArtificial AnalysisArtificial AnalysisArtificial AnalysisArtificial AnalysisArtificial Analysis

Sep 2, 2026

Meta Reaches the Frontier, and Keeps It Closed

Meta released Muse Spark 1.3, its fourth in five months. On the Artificial Analysis Intelligence Index the limited-preview 'max' version scores 62 — behind only Claude Fable 5.1 and Claude Opus 5 — and the generally available version scores 61, the cheapest model at that level ($0.55 per task, unchanged $1.25/$4.25 pricing). It is Meta's strongest showing since it made open weights a strategy, but the model is proprietary, served through Meta's own API with the top version limited to partners. Its August pledge to 'soon' open Muse Spark 1.2's weights remains unmet.

Read

Artificial Analysis

Sep 2, 2026

Six Frontier-Band Models in a Single Week

Between 27 August and 3 September, five labs shipped models scoring 57 to 66 on the Artificial Analysis Intelligence Index: Anthropic's Claude Fable 5.1 (66), Meta's Muse Spark 1.3 (62), OpenAI's GPT-6 Astra (61), Google's Gemini 3.8 Flash (59, its fourth Flash model in under four months), and Zhipu's open-weight GLM-5.3-Flash (57, MIT-licensed, emerging from a week-long anonymous 'Ox Alpha' preview). Tencent's open-weight Hy4 arrived the same week, still unscored — a sixth release, from a sixth lab. Every one of the top scores belongs to a closed model.

Read

Artificial AnalysisArtificial AnalysisArtificial AnalysisQubrid AI

Sep 1, 2026

Claude Fable 5.1 Sets the Highest Score Yet, and the Open Gap Doubles

Anthropic released Claude Fable 5.1, and its trusted-access sibling Mythos 5.1, and on the independent Artificial Analysis Intelligence Index its top effort scores 66 — the highest the index has measured, four points above Fable 5 and three above Claude Opus 5. Standard token pricing is unchanged at $10/$50 per million, but Anthropic cut the cache-read price 75%, from $1 to $0.25; even so, a task costs about 20% more than on Fable 5, because the model spends roughly 1.7 times the output tokens to reach the higher score. The move reopens the distance to the downloadable frontier: the best open-weight models, Moonshot's Kimi K3 and Zhipu's GLM-5.3, sit at 60, so the gap between the best model and the best one anyone can download widened from three points to six in a single release.

Read

AnthropicArtificial Analysis

Aug 24, 2026

A Publisher Says It Reached the Frontier for $40 Million

Thomson Reuters launched Thomson, its own large language model, built by taking an open-source foundation and spending $40 million on talent and compute to mid-train and post-train it over Westlaw, Practical Law, Checkpoint and Reuters — less than 10 percent of the company's content so far. It says the model is fully owned in-house and runs without the inference costs of a typical frontier model, and its chief executive says early evaluations put it "on par with the latest frontier models across a range of tasks". No benchmark is named, no score published and no third party has assessed it, so the capability claim is the company's alone.

Read

Thomson Reuters (corporate)

Our reading

Aug 17, 2026

Three of the four biggest labs now publish a price with a date or an hour attached

Google's pricing page lists Gemini 3.7 Flash and Gemini 3.6 Flash at $0.75 per million input tokens and $3.75 per million output "through December 31, 2026", and at $1.50 and $7.50 "starting January 1, 2027". DeepSeek's price table halves its rates only outside two daily peak windows, so the same request costs two different amounts depending on the hour. Anthropic's documentation, which in July carried two dated rows for Claude Sonnet 5, now carries one at $2 and $10 with no successor row — the company withdrew its scheduled increase rather than adding one. OpenAI's pricing page lists gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna at $5/$30, $2/$12 and $0.20/$1.20, with no dated future rate anywhere on the page. On two of the four schedules the rate a buyer pays today and the rate already published for next year differ by a factor of two.

Compared

Google DeepMindDeepSeekAnthropicAnthropicOpenAI

Aug 16, 2026

The cheapest model on the board raises its prices several-fold

DeepSeek's new prices took effect at 16:00 UTC on 16 August, replacing flat rates with a peak and off-peak schedule. V4-Flash, priced through July at $0.14 per million input tokens and $0.28 per million output, now costs $0.44 and $1.32 in peak hours and half that outside them. V4-Pro moves from $0.435 and $0.87 to $1.32 and $3.96 at peak, $0.66 and $1.98 off-peak. Reported increases run from 50% to more than 1,100% depending on the model, the token type and the hour. Peak covers seven hours of every 24 — 01:00 to 04:00 and 06:00 to 10:00 UTC — so most of the day bills at the lower rate. The company said it was adopting the schedule "to allocate resources more reasonably ... encouraging users to schedule their tasks based on actual usage", announcing it alongside the general release of V4-Pro. Both models' weights remain published under the MIT licence, so the rise applies to DeepSeek's hosted service rather than to the models, which anyone can still run themselves. Even at peak, V4-Pro's output tokens cost about a twelfth of Claude Fable 5's $50. Four months earlier DeepSeek had cut V4-Pro by 75% and its cached-input prices across the API to a tenth of their previous level.

Read

DeepSeekDeepSeekYahoo FinanceFortuneArtificial AnalysisArtificial Analysis

Aug 13, 2026

Google ships Gemini 3.7 Flash three weeks after 3.6, at half price until the new year

Google released Gemini 3.7 Flash 23 days after Gemini 3.6 Flash, priced at $0.75 per million input tokens and $3.75 per million output through 31 December 2026, with $1.50 and $7.50 taking effect on 1 January 2027. It cut 3.6 Flash to the same dated rate, so the new model does not undercut its predecessor — both undercut the price 3.6 Flash launched at three weeks earlier. On the independent Artificial Analysis Intelligence Index 3.7 Flash scores 56, four points above 3.6 Flash and level with xAI's Grok 4.5 at a third of Grok's input price. Google's own benchmark table claims leads on production-code quality and web development; the same table shows OpenAI's GPT-5.6 Terra ahead on two terminal benchmarks and Claude Sonnet 5 ahead on a desktop-agent evaluation, and those are the lab's own figures either way. Google gave no release date for its next flagship, Gemini 3.5 Pro, which it had previously described as in partner testing.

Read

Google DeepMindArtificial AnalysisVentureBeat

Aug 10, 2026

Anthropic cancels the price rise it had already scheduled

Anthropic withdrew the increase it had published for Claude Sonnet 5 when the model shipped on 30 June, adding an editor's note to the launch announcement: "Sonnet 5's introductory pricing of $2 per million input tokens and $10 per million output tokens is now permanent. The standard pricing of $3 input / $15 output previously set to take effect September 1 no longer applies." Its pricing documentation, which through July carried two dated rows for the model, now lists a single row at $2 and $10. Sonnet 5 is the default model on Anthropic's free and paid consumer plans, and the withdrawn increase was the only scheduled rise on any of the three largest American labs' published price lists.

Read

AnthropicAnthropicPYMNTS

Aug 5, 2026

Meta ships Muse Spark 1.2, its third model in four months

Meta released Muse Spark 1.2, a proprietary model scoring 54 on the Artificial Analysis Intelligence Index v4.1 — three points above Muse Spark 1.1's 51, inside the frontier band but below the leaders. It is priced at $1.25 per million input tokens and $4.25 per million output, unchanged from Muse Spark 1.1, with a one-million-token context window and text-and-image input, and is optimized for coding. Like its predecessor it is closed: Artificial Analysis lists the weights as not publicly available and Meta discloses no parameter count. It is Meta's third model release in four months, from the company whose Llama line made open-weight frontier-chasing a corporate strategy before its current flagship range went proprietary.

Read

Artificial AnalysisArtificial Analysis

Aug 3, 2026

Alibaba releases Qwen3.8-Max, a proprietary frontier-band model

Alibaba released Qwen3.8-Max, a 2.4-trillion-parameter sparse mixture-of-experts model scoring 56 on the Artificial Analysis Intelligence Index v4.1 — inside the frontier band, below Moonshot's Kimi K3 (57), OpenAI's GPT-5.6 Sol (59) and Anthropic's Claude Opus 5 (61), and above Grok 4.5, Claude Sonnet 5 and GPT-5.5. It is priced at $2.00 per million input tokens and $6.00 per million output, with a one-million-token context window. Artificial Analysis classifies the model as proprietary — available through Alibaba's API, weights not published. Stanford's HAI and some trade coverage described Qwen3.8-Max as open-weight, and Alibaba separately released a smaller 27-billion-parameter open-weight variant, but the 2.4-trillion Max that carries the index score is API-only on the independent index's classification. Alibaba's own blog said the weights would be released; at launch they had not been, so on the measurement this record uses the largest new Chinese model counts as closed.

Read

Alibaba (Qwen team)Artificial Analysis

Our reading

Jul 30, 2026

A frontier price was scheduled to rise by half, and the token it is charged on has changed size

The published direction of travel was not all downward. Anthropic's pricing documentation carried two dated rows for Claude Sonnet 5: $2.00 and $10.00 per million input and output tokens "through August 31, 2026", then $3.00 and $15.00 "starting September 1, 2026" — a 50% increase on both sides, published in advance since the model shipped on 30 June rather than sprung, and applying to the model Anthropic made the default in its Free and Pro plans. The increase was withdrawn on 10 August and the $2/$10 rate made permanent; the same documentation now carries a single Sonnet 5 row. The second thing it disclosed still stands, and it cuts across every per-token comparison here: models from Claude 4.7 onward use a newer tokenizer producing roughly 30% more tokens for the same text, so a price per token is not the same unit across model generations. Every other Anthropic price was re-checked against the same page and is unchanged: Claude Fable 5 at $10/$50, Claude Opus 5 and Claude Opus 4.8 at $5/$25.

Compared

AnthropicAnthropicAnthropic

Jul 30, 2026

DeepMind sends Gemini into the physical world as a robot's planner

Google DeepMind released Gemini Robotics ER 2, an embodied-reasoning model that acts as a robot's high-level planner — reading continuous video, tracking its own task progress, coordinating multiple robots — while delegating motor control to lower-level vision-language-action models. It ships publicly through the Gemini API rather than as a research demo, which is the substantive step: frontier-lab reasoning models are now a purchasable component for physical machines, with whatever that implies for the deployment questions this record tracks mostly in software.

Read

Google DeepMind

Jul 30, 2026

OpenAI cuts GPT-5.6 Luna 80% in a frontier price war

OpenAI cut the API price of two GPT-5.6 models, attributing the move to efficiency gains passed to customers rather than a promotion. Luna, the series' fast model, fell 80% to $0.20 per million input tokens and $1.20 per million output, from $1/$6; Terra, the everyday model, fell 20% to $2/$12 from $2.50/$15. The flagship, Sol, held at $5/$30 but gained a "Sol Fast" mode at twice standard price ($10/$60) for 2.5× throughput with no change in intelligence, replacing Priority Processing. On the Artificial Analysis Intelligence Index v4.1 Luna scores 51 and Terra 55; at the new price Artificial Analysis puts Luna at about $0.21 to run its test battery, undercutting Zhipu's GLM-5.2 as the cheapest model at that intelligence level. The cut is the latest in a run of releases that, by Artificial Analysis's count, put six labs above 50 on that index within eight days — the price of near-frontier capability being competed down rather than a single vendor's discount.

Read

OpenAICNBCVentureBeatArtificial Analysis

Earlier this year

Jul 27, 2026“It is the world's first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning.” — Moonshot AI✓Fact check: true
Jul 27, 2026Microsoft ships a security model that cannot be evaluated from outside
Jul 27, 2026“This combined expertise delivers exceptional security protection, beating Mythos, Gemini and GPT on CyberGym, the gold standard benchmark for evaluating how systems reason over large codebases to find real vulnerabilities in the code.” — Microsoft±Fact check: mixed
Jul 27, 2026Three cyber-specialist releases in seven days, none of them priced or independently scored
Jul 27, 2026“Distillation does not allow the CCP to obtain equivalent or superior AI capabilities to the US, but it can bring the Chinese frontier to within a few months of the US frontier.” — Dario Amodei±Fact check: mixed
Jul 24, 2026Anthropic releases Claude Opus 5 at half of Fable 5’s per-token price
Jul 24, 2026“Claude Opus 5 is available today. It's a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price.” — Anthropic±Fact check: mixed
Jul 23, 2026OpenAI previews GPT-5.6 to trusted partners first, at Washington's request
Jul 22, 2026“We are creating a baseline on top of which you will see us rapidly iterate on subsequent model releases. And so picking up pace and releasing models almost at a monthly cadence is part of our road map as we are building Gemini 4 as well,” — Sundar Pichai
Jul 21, 2026Google ships three Gemini models — and not the flagship
Jul 21, 2026Google restricts its new vulnerability-hunting model to governments and trusted partners
Jul 21, 2026“Gemini 3.5 Pro is currently testing with partners and we plan to make it broadly available as soon as it's ready.” — Google DeepMind
Jul 16, 2026xAI launches Grok 4.5 on token efficiency, not peak capability
Jul 9, 2026OpenAI rolls out the GPT-5.6 series: Sol, Terra, and Luna
Jun 30, 2026Export controls on Fable 5 and Mythos 5 lifted; access restored
Jun 30, 2026Claude Sonnet 5 becomes Anthropic's default model
Jun 26, 2026OpenAI limits GPT-5.6 models to trusted partners at US government request
Jun 12, 2026
US government applies export controls to Claude Fable 5 and Mythos 5An unprecedented intervention: export controls on specific commercial AI models, requiring nationality-based restriction. Unable to verify nationality in-product, Anthropic suspended access rather than operate out of compliance — the first time a frontier model was pulled from the market by government order.
Read
Anthropicanthropic.com61% credibilityEarned from 4 checked claims.Anthropicanthropic.com61% credibilityEarned from 4 checked claims.Anthropicanthropic.com61% credibilityEarned from 4 checked claims.CNBCcnbc.com80% credibility · not yet testedAn opening estimate from its institutional history — none of its claims checked yet.
Jun 9, 2026Anthropic launches Claude Fable 5 and Mythos 5, a tier above Opus
Jun 8, 2026“Our internal belief is that by March of 2028 we may have a significant fraction of our research being done by AI systems in tandem with our own researchers.” — OpenAI
May 19, 2026Google opens the Gemini 3.5 generation with Flash
Apr 23, 2026GPT-5.5 arrives with OpenAI's first dual 'High' risk designation
Feb 13, 2026OpenAI retires GPT-4o, GPT-4.1, o4-mini, and GPT-5 from ChatGPT
Nov 24, 2025Anthropic's Claude Opus 4.5 caps three major launches in two months
Nov 18, 2025Google ships Gemini 3, embedding it in Search on day one

Earlier

Aug 7, 2025OpenAI releases GPT-5 to all ChatGPT users
Aug 7, 2025“This is the best model in the world at coding. This is the best model in the world at writing, the best model in the world at health care, and a long list of things beyond that.” — Sam Altman
Aug 4, 2025“It's a very good thing for the open community.” — Nathan Lambert✓Fact check: true
Jul 9, 2025xAI releases Grok 4, claiming the frontier's top spot
Jul 9, 2025“Grok 4 is the most intelligent model in the world. It includes native tool use and real-time search integration.” — xAI
May 22, 2025Anthropic releases Claude Opus 4 and Sonnet 4
Apr 28, 2025“The U.S. is doubling down on restricting sales of chips to China and purchases from China, but models like Qwen 3 that are state-of-the-art and open […] will undoubtedly be used domestically. It reflects the reality that businesses are both building their own tools [as well as] buying off the shelf via closed-model companies like Anthropic and OpenAI.” — Tuhin Srivastava✓Fact check: true
Apr 16, 2025OpenAI's o3 and o4-mini make reasoning models agentic
Apr 5, 2025Meta's Llama 4 launch is marred by a benchmark-integrity controversy
Mar 25, 2025Google's Gemini 2.5 Pro takes the LMArena lead
Feb 19, 2025xAI ships Grok 3 with reasoning variants and a million-token context
Sep 12, 2024OpenAI's o1 opens the reasoning-model era
Jul 18, 2024GPT-4o mini resets the price floor at $0.15 per million tokens
Jun 20, 2024Claude 3.5 Sonnet delivers Opus-class capability at a fifth of the price
Jun 10, 2024Apple Intelligence puts a ~3B-parameter model on the iPhone
May 13, 2024OpenAI releases GPT-4o and brings frontier capability to the free tier
Mar 4, 2024Anthropic releases the Claude 3 family
Mar 4, 2024“Claude 3 Opus has self-reported benchmark scores that consistently beat GPT-4. This is a really big deal: in the 12+ months since the GPT-4 release no other model has consistently beat it in this way.” — Simon Willison✓Fact check: true
Feb 15, 2024Google's Gemini 1.5 Pro debuts the million-token context window
Dec 6, 2023Google announces Gemini, its first natively multimodal model family
Dec 6, 2023“I think we're substantially ahead on 30 out of 32” — Demis Hassabis±Fact check: mixed
Mar 14, 2023OpenAI releases GPT-4
Mar 14, 2023Anthropic opens access to Claude
Mar 14, 2023“it passes a simulated bar exam with a score around the top 10% of test takers; in contrast, GPT-3.5's score was around the bottom 10%.” — OpenAI
Nov 30, 2022
OpenAI releases ChatGPT as a free research previewA conversational model fine-tuned with RLHF from a GPT-3.5-series base went live to the public at an effective price of $0. OpenAI's own launch notes disclosed the system 'sometimes writes plausible-sounding but incorrect or nonsensical answers.' The release turned large language models from a developer API into a mass consumer phenomenon and set off the competitive race that defines this page.
Read
OpenAIopenai.com65% credibilityEarned from 5 checked claims.
Apr 12, 2022DeepMind's Chinchilla finds the era's giant models undertrained
Jun 11, 2020OpenAI opens the API era with GPT-3

Covered since Jun 2020Read it from the start →